# Understanding Non-equilibrium

Thermodynamics

G. Lebon • D. Jou • J. Casas-Vázquez

Understanding Non-equilibrium

Thermodynamics

Foundations, Applications, Frontiers

123

Prof. Dr. David Jou

Universitat Autònoma de Barcelona

Dept. Fisica - Ediﬁci Cc

Grup Fisica Estadistica

Bellaterra 08193

Catalonia, Spain

David.Jou@uab.es

Prof. Dr. José Casas-Vázquez

Universitat Autònoma de Barcelona

Dept. Fisica - Ediﬁci Cc

Grup Fisica Estadistica

Bellaterra 08193

Catalonia, Spain

Jose.Casas@uab.es

Prof. Dr. Georgy Lebon

Université de Liège

Dept. d’Astrophysique

G´ eophysique et Oc´ eanographie

B5 Sart Tilman

Liege 4000

Belgium

g.lebon@ulg.ac.be

ISBN: 978-3-540-74251-7 e-ISBN: 978-3-540-74252-4

Library of Congress Control Number: 2007935452

c 2008 Springer-Verlag Berlin Heidelberg

This work is subject to copyright. All rights are reserved, whether the whole or part of the material is

concerned, speciﬁcally the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting,

reproduction on microﬁlm or in any other way, and storage in data banks. Duplication of this publication

or parts thereof is permitted only under the provisions of the German Copyright Law of September 9,

1965, in its current version, and permission for use must always be obtained from Springer. Violations are

liable to prosecution under the German Copyright Law.

The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply,

even in the absence of a speciﬁc statement, that such names are exempt from the relevant protective laws

and regulations and therefore free for general use.

Cover design: WMX Design GmbH, Heidelberg

Printed on acid-free paper

9 8 7 6 5 4 3 2 1

springer.com

Cover image: Thermal radiation leaving the Earth, seen by the EOS-Terra satellite

(NASA). Image from www.visibleearth.nasa.gov. Owner: NASA.

Preface

Our time is characterized by an explosion of information and by an acceler-

ation of knowledge. A book cannot compete with the huge amount of data

available on the Web. However, to assimilate all this information, it is nec-

essary to structure our knowledge in a useful conceptual framework. The

purpose of the present work is to provide such a structure for students and

researchers interested by the current state of the art of non-equilibrium ther-

modynamics. The main features of the book are a concise and critical presen-

tation of the basic ideas, illustrated by a series of examples, selected not only

for their pedagogical value but also for the perspectives oﬀered by recent

technological advances. This book is aimed at students and researchers in

physics, chemistry, engineering, material sciences, and biology.

We have been guided by two apparently antagonistic objectives: general-

ity and simplicity. To make the book accessible to a large audience of non-

specialists, we have decided about a simpliﬁed but rigorous presentation.

Emphasis is put on the underlying physical background without sacriﬁcing

mathematical rigour, the several formalisms being illustrated by a list of ex-

amples and problems. All over this work, we have been guided by the formula:

“Get the more from the less”, with the purpose to make a maximum of people

aware of a maximum of knowledge from a minimum of basic tools.

Besides being an introductory text, our objective is to present an overview,

as general as possible, of the more recent developments in non-equilibrium

thermodynamics, especially beyond the local equilibrium description. This

is partially a terra incognita, an unknown land, because basic concepts as

temperature, entropy, and the validity of the second law become problematic

beyond the local equilibrium hypothesis. The answers provided up to now

must be considered as partial and provisional, but are nevertheless worth to

be examined.

Chapters 1 and 2 are introductory chapters in which the main concepts

underlying equilibrium thermodynamics and classical non-equilibrium ther-

modynamics are stated. The basic notions are discussed with special emphasis

on these needed later in this book.

V

VI Preface

Several applications of classical non-equilibrium thermodynamics are pre-

sented in Chaps. 3 and 4. These illustrations have not been chosen arbitrarily,

but keeping in mind the perspectives opened by recent technological advance-

ments. For instance, advances in material sciences have led to promising

possibilities for thermoelectric devices; localized intense laser heating used to

make easier the separation of molecules has contributed to a revival of inter-

est in thermodiﬀusion; chemical reactions are of special interest in biology,

in relation with their coupling with active transport across membranes and

recent developments of molecular motors.

The purpose of Chaps. 5 and 6 is to discuss two particular aspects of

classical non-equilibrium thermodynamics which have been the subject of

active research during the last decades. Chapter 5 is devoted to ﬁnite-time

thermodynamics whose main concern is the competition between maximum

eﬃciency and maximum power and its impact on economy and ecology. This

classical subject is treated here in an updated form, taking into account the

last technological possibilities and challenges, as well as some social con-

cerns. Chapter 6 deals with instabilities and pattern formation; organized

structures occur in closed and open systems as a consequence of ﬂuctuations

growing far from equilibrium under the action of external forces. Patterns are

observed in a multitude of our daily life experiences, like in hydrodynamics,

biology, chemistry, electricity, material sciences, or geology. After introducing

the mathematical theory of stability, several examples of ordered structures

are analysed with a special attention to the celebrated B´enard cells.

Chapters 1–6 may provide a self-consistent basis for a graduate introduc-

tory course in non-equilibrium thermodynamics.

In the remainder of the book, we go beyond the framework of the classical

description and spend some time to address and compare the most recent

developments in non-equilibrium thermodynamics. Chapters 7–11 will be of

interest for students and researchers, who feel attracted by new scientiﬁc

projects wherein they may be involved. This second part of the book may

provide the basis for an advanced graduate or even postgraduate course on

the several trends in contemporary thermodynamics.

The coexistence of several schools in non-equilibrium thermodynamics is

a reality; it is not a surprise in view of the complexity of most macroscopic

systems and the fact that some basic notions as temperature and entropy are

not univocally deﬁned outside equilibrium. To appreciate this form of multi-

culturalism in a positive sense, it is obviously necessary to know what are the

foundations of these theories and to which extent they are related. A superﬁ-

cial inspection reveals that some viewpoints are overlapping but none of them

is rigorously equivalent to the other. A detailed and complete understanding

of the relationship among the diverse schools turns out to be not an easy

task. The ﬁrst diﬃculty stems from the fact that each approach is associated

with a certain insight, we may even say an intuition or feeling that is some-

times rather diﬃcult to apprehend. Also some unavoidable diﬀerences in the

terminology and the notation do not facilitate the communication. Another

Preface VII

factor that contributes to the diﬃculty to reaching a mutual comprehension

is that the schools are not frozen in time: they evolve as a consequence of

internal dynamics and by contact with others. Our goal is to contribute to

a better understanding among the diﬀerent schools by discussing their main

concepts, results, advantages, and limitations. Comparison of diﬀerent view-

points may be helpful for a deeper comprehension and a possible synthesis of

the many faces of the theory. Such a comparative study is not found in other

textbooks.

One problem was the selection of the main representative ones among the

wealth of thermodynamic formalisms. Here we have focused our attention

on ﬁve of them: extended thermodynamics (Chap. 7), theories with internal

variables (Chap. 8), rational thermodynamics (Chap. 9), Hamiltonian formu-

lation (Chap. 10), and mesoscopic approaches (Chap.11). In each of them, we

have tried to save the particular spirit of each theory.

It is clear that our choice is subjective: we have nevertheless been guided

not only by the pedagogical aspect and/or the impact and universality of the

diﬀerent formalisms, but also by the fact that we had to restrict ourselves.

Moreover, it is our belief that a good comprehension of these diﬀerent ver-

sions allows for a better and more understandable comprehension of theories

whose opportunity was not oﬀered to be discussed here. The common points

shared by the theories presented in Chaps. 7–11 are not only to get rid of the

local equilibrium hypothesis, which is the pillar of the classical theory, but

also to propose new phenomenological approaches involving non-linearities,

memory and non-local eﬀects, with the purpose to account for the techno-

logical requirements of faster processes and more miniaturized devices.

It could be surprising that the book is completely devoted to macroscopic

and mesoscopic aspects and that microscopic theories have been widely omit-

ted. The reasons are that many excellent treatises have been written on mi-

croscopic theories and that we decided to keep the volume of the book to a

reasonable ratio. Although statistical mechanics appears to be more fashion-

able than thermodynamics in the eyes of some people and the developments

of microscopic methods are challenging, we hope to convince the reader that

macroscopic approaches, like thermodynamics, deserve a careful attention

and are the seeds of the progress of knowledge. Notwithstanding, we remain

convinced that, within the perspectives of improvement and uniﬁcation, it is

highly desirable to include as many microscopic results as possible into the

macroscopic framework.

Chapters 7–11 are autonomous and self-consistent, they have been struc-

tured in such a way that they can be read independently of each other and

in arbitrary order. However, it is highly recommended to browse through all

the chapters to better apprehend the essence and the complementarity of the

diverse theories.

At the end of each chapter is given a list of problems. The aim is not

only to allow the reader to check his understanding, but also to stimulate

his interest to solve concrete situations. Some of these problems have been

VIII Preface

inspired by recent papers, which are mentioned, and which may be consulted

for further investigation. More technical and advanced parts are conﬁned in

boxes and can be omitted during a ﬁrst reading.

We acknowledge many colleagues, and in particular M. Grmela (Montreal

University), P.C Dauby and Th. Desaive (Li`ege University), for the discus-

sions on these and related topics for more than 30 years. We also appreciate

our close collaborators for their help and stimulus in research and teach-

ing. Drs. Vicen¸ c M´endez and Vicente Ortega-Cejas deserve special gratitude

for their help in the technical preparation of this book. We also acknowl-

edge the ﬁnantial support of the Direcci´ on General de Investigaci´ on of the

Spanish Ministry of Education under grants BFM2003-06003 and FIS2006-

12296-C02-01, and of the Direcci´ o General de Recerca of the Generalitat of

Catalonia, under grants 2001 SGR 00186 and 2005 SGR 00087.

Li`ege-Bellaterra, March 2007

G. Lebon,

D. Jou,

J. Casas-V´azquez

Contents

1 Equilibrium Thermodynamics: A Review . . . . . . . . . . . . . . . . . 1

1.1 The Early History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Scope and Deﬁnitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 The Fundamental Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.3.1 The Zeroth Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.3.2 The First Law or Energy Balance . . . . . . . . . . . . . . . . . . 6

1.3.3 The Second Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.3.4 The Third Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.4 Gibbs’ Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.4.1 Fundamental Relations and State Equations . . . . . . . . . 15

1.4.2 Euler’s Relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.4.3 Gibbs–Duhem’s Relation . . . . . . . . . . . . . . . . . . . . . . . . . . 16

1.4.4 Some Deﬁnitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

1.4.5 The Basic Problem of Equilibrium Thermodynamics . . 18

1.5 Legendre Transformations and Thermodynamic Potentials . . . 19

1.5.1 Thermodynamic Potentials . . . . . . . . . . . . . . . . . . . . . . . . 20

1.5.2 Thermodynamic Potentials and Extremum Principles . 21

1.6 Stability of Equilibrium States . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

1.6.1 Stability of Single Component Systems . . . . . . . . . . . . . . 24

1.6.2 Stability Conditions for the Other Thermodynamic

Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

1.6.3 Stability Criterion of Multi-Component Mixtures . . . . . 27

1.7 Equilibrium Chemical Thermodynamics . . . . . . . . . . . . . . . . . . . 29

1.7.1 General Equilibrium Conditions . . . . . . . . . . . . . . . . . . . . 30

1.7.2 Heat of Reaction and van’t Hoﬀ Relation . . . . . . . . . . . . 31

1.7.3 Stability of Chemical Equilibrium and Le Chatelier’s

Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

1.8 Final Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

1.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

IX

X Contents

2 Classical Irreversible Thermodynamics . . . . . . . . . . . . . . . . . . . 37

2.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.2 Local Equilibrium Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

2.3 Entropy Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

2.4 General Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

2.5 Stationary States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

2.5.1 Minimum Entropy Production Principle . . . . . . . . . . . . . 51

2.6 Applications to Heat Conduction, Mass Transport,

and Fluid Flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

2.6.1 Heat Conduction in a Rigid Body . . . . . . . . . . . . . . . . . . 54

2.6.2 Matter Diﬀusion Under Isothermal and Isobaric

Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

2.6.3 Hydrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

2.7 Limitations of the Classical Theory of Irreversible

Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

2.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

3 Coupled Transport Phenomena . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.1 Electrical Conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

3.2 Thermoelectric Eﬀects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.2.1 Phenomenological Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.2.2 Eﬃciency of Thermoelectric Generators . . . . . . . . . . . . . 76

3.3 Thermodiﬀusion: Coupling of Heat and Mass Transport . . . . . 79

3.4 Diﬀusion Through a Membrane . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.4.1 Entropy Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.4.2 Phenomenological Relations . . . . . . . . . . . . . . . . . . . . . . . 85

3.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

4 Chemical Reactions and Molecular Machines . . . . . . . . . . . . . 91

4.1 One Single Chemical Reaction . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

4.2 Coupled Chemical Reactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

4.2.1 General Formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

4.2.2 Cyclical Chemical Reactions and Onsager’s

Reciprocal Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

4.3 Eﬃciency of Energy Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

4.4 Chemical Reactions and Mass Transport:

Molecular Machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

4.5 Autocatalytic Reactions and Diﬀusion: Morphogenesis . . . . . . 108

4.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

5 Finite-Time Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

5.1 The Finite-Time Carnot Cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

5.1.1 Curzon–Ahlborn’s Model: Heat Losses . . . . . . . . . . . . . . 115

5.1.2 Friction Losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

5.2 Economical and Ecological Constraints . . . . . . . . . . . . . . . . . . . . 122

Contents XI

5.3 Earth’s Atmosphere as a Non-Equilibrium System

and a Heat Engine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

5.3.1 Earth’s Energy Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

5.3.2 Global Warming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

5.3.3 Transformation of Solar Heat into Wind Motion . . . . . . 128

5.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

6 Instabilities and Pattern Formation . . . . . . . . . . . . . . . . . . . . . . . 135

6.1 The Linear Theory of Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

6.2 Non-Linear Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

6.3 Thermal Convection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

6.3.1 The Rayleigh–B´enard’s Instability: A Linear Theory . . 145

6.3.2 The Rayleigh–B´enard’s Instability:

A Non-Linear Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

6.3.3 B´enard–Marangoni’s Surface Tension-Driven

Instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

6.4 Taylor’s Instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

6.5 Chemical Instabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

6.5.1 Temporal Organization in Spatially Homogeneous

Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

6.5.2 Spatial Organization in Spatially Heterogeneous

Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

6.5.3 Spatio-Temporal Patterns in Heterogeneous Systems:

Turing Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

6.6 Miscellaneous Examples of Pattern Formation . . . . . . . . . . . . . . 169

6.6.1 Salt Fingers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

6.6.2 Patterns in Electricity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171

6.6.3 Dendritic Pattern Formation . . . . . . . . . . . . . . . . . . . . . . . 172

6.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

7 Extended Irreversible Thermodynamics. . . . . . . . . . . . . . . . . . . 179

7.1 Heat Conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

7.1.1 Fourier’s vs. Cattaneo’s Law . . . . . . . . . . . . . . . . . . . . . . . 181

7.1.2 Extended Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

7.1.3 Non-Local Terms: From Collision-Dominated Regime

to Ballistic Regime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191

7.1.4 Application to Steady Heat Transport

in Nano-Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

7.2 One-Component Viscous Heat Conducting Fluids . . . . . . . . . . . 196

7.3 Rheological Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200

7.4 Microelectronic Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202

7.5 Final Comments and Perspectives . . . . . . . . . . . . . . . . . . . . . . . . 205

7.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209

XII Contents

8 Theories with Internal Variables . . . . . . . . . . . . . . . . . . . . . . . . . . 215

8.1 General Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216

8.1.1 Accompanying State Axiom . . . . . . . . . . . . . . . . . . . . . . . 216

8.1.2 Entropy and Entropy Production . . . . . . . . . . . . . . . . . . . 219

8.1.3 Rate Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220

8.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

8.2.1 Viscoelastic Solids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221

8.2.2 Polymeric Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224

8.2.3 Colloidal Suspensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227

8.3 Final Comments and Comparison with Other Theories . . . . . . 232

8.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234

9 Rational Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237

9.1 General Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238

9.2 The Axioms of Rational Thermodynamics . . . . . . . . . . . . . . . . . 238

9.2.1 Axiom of Admissibility and Clausius–Duhem’s

Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239

9.2.2 Axiom of Memory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240

9.2.3 Axiom of Equipresence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241

9.2.4 Axiom of Local Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241

9.2.5 Axiom of Material Frame-Indiﬀerence . . . . . . . . . . . . . . . 242

9.3 Application to Thermoelastic Materials . . . . . . . . . . . . . . . . . . . 243

9.4 Viscous Heat Conducting Fluids . . . . . . . . . . . . . . . . . . . . . . . . . . 247

9.5 Comments and Critical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 249

9.5.1 The Clausius–Duhem’s Inequality . . . . . . . . . . . . . . . . . . 249

9.5.2 Axiom of Phlogiston . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249

9.5.3 The Meaning of Temperature and Entropy . . . . . . . . . . 250

9.5.4 Axiom of Frame-Indiﬀerence . . . . . . . . . . . . . . . . . . . . . . . 251

9.5.5 The Entropy Flux Axiom. . . . . . . . . . . . . . . . . . . . . . . . . . 252

9.5.6 The Axiom of Equipresence . . . . . . . . . . . . . . . . . . . . . . . . 252

9.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257

10 Hamiltonian Formalisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261

10.1 Classical Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262

10.2 Formulation of GENERIC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264

10.2.1 Classical Navier–Stokes’ Hydrodynamics . . . . . . . . . . . . 266

10.2.2 Fickian Diﬀusion in Binary Mixtures . . . . . . . . . . . . . . . . 270

10.2.3 Non-Fickian Diﬀusion in Binary Mixtures . . . . . . . . . . . 273

10.3 Final Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274

10.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278

11 Mesoscopic Thermodynamic Descriptions . . . . . . . . . . . . . . . . . 279

11.1 Einstein’s Formula: Second Moments

of Equilibrium Fluctuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279

11.2 Derivation of the Onsager–Casimir’s Reciprocal Relations . . . . 282

Contents XIII

11.3 Fluctuation–Dissipation Theorem. . . . . . . . . . . . . . . . . . . . . . . . . 285

11.4 Keizer’s Theory: Fluctuations in Non-Equilibrium Steady

States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288

11.4.1 Dynamics of Fluctuations . . . . . . . . . . . . . . . . . . . . . . . . . 288

11.4.2 A Non-Equilibrium Entropy . . . . . . . . . . . . . . . . . . . . . . . 289

11.5 Mesoscopic Non-Equilibrium Thermodynamics . . . . . . . . . . . . . 292

11.5.1 Brownian Motion with Inertia. . . . . . . . . . . . . . . . . . . . . . 293

11.5.2 Other Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296

11.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299

Epilogue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307

Further Readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321

Chapter 1

Equilibrium Thermodynamics: A Review

Equilibrium States, Reversible Processes,

Energy Conversion

Equilibrium or classical thermodynamics deals essentially with the study of

macroscopic properties of matter at equilibrium. A comprehensive deﬁnition

of equilibrium will be given later; here it is suﬃcient to characterize it as a

time-independent state, like a column of air at rest in absence of any ﬂux of

matter, energy, charge, or momentum. By extension, equilibrium thermody-

namics has also been applied to the description of reversible processes: they

represent a special class of idealized processes considered as a continuum

sequence of equilibrium states.

Since time does not appear explicitly in the formalism, it would be more

appropriate to call it thermostatics and to reserve the name thermodynamics

to the study of processes taking place in the course of time outside equilib-

rium. However, for historical reasons, the name “thermodynamics” is widely

utilized nowadays, even when referring to equilibrium situations. We shall

here follow the attitude dictated by the majority but, to avoid any confu-

sion, we shall speak about equilibrium thermodynamics and designate beyond-

equilibrium theories under the name of non-equilibrium thermodynamics.

The reader is assumed to be already acquainted with equilibrium thermo-

dynamics but, for the sake of completeness, we brieﬂy recall here the essential

concepts needed along this book. This chapter will run as follows. After a

short historical introduction and a brief recall of basic deﬁnitions, we present

the fundamental laws underlying equilibrium thermodynamics. We shall put

emphasis on Gibbs’ equation and its consequences. After having established

the criteria of stability of equilibrium, a last section, will be devoted to an

introduction to chemical thermodynamics.

1.1 The Early History

Equilibrium thermodynamics is the natural extension of the older science,

Mechanics. The latter, which rests on Newton’s law, is essentially concerned

with the study of motions of idealized systems as mass-particles and rigid

1

2 1 Equilibrium Thermodynamics: A Review

solids. Two important notions, heat and temperature, which are absent in

mechanics, constitute the pillars of the establishment of equilibrium ther-

modynamics as a branch of science. The need to develop a science beyond

the abstract approach of Newton’s law to cope with the reality of engi-

neer’s activities was born in the beginning of nineteenth century. The ﬁrst

steps and concepts of thermodynamics were established by Fourier, Carnot,

Kelvin, Clausius, and Gibbs among others. Thermodynamics began in 1822

with Fourier’s publication of the Th´eorie analytique de la chaleur wherein

is derived the partial diﬀerential equation for the temperature distribution

in a rigid body. Two years later, in 1824, Sadi Carnot (1796–1832) put

down further the foundations of thermodynamics with his renowned mem-

oir R´eﬂexions sur la puissance motrice du feu et sur les machines propres ` a

d´evelopper cette puissance. Carnot perceived that steam power was a motor

of industrial revolution that would prompt economical and social life. Al-

though a cornerstone in the formulation of thermodynamics, Carnot’s work

is based on several misconceptions, as for instance the identiﬁcation of heat

with a hypothetical indestructible weightless substance, the caloric, a notion

introduced by Lavoisier. Signiﬁcant progresses towards a better comprehen-

sion of the subject can be attributed to a generation of outstanding scientists

as James P. Joule (1818–1889) who identiﬁed heat as a form of energy trans-

fer by showing experimentally that heat and work are mutually convertible.

This was the birth of the concept of energy and the basis of the formulation

of the ﬁrst law of thermodynamics. At the same period, William Thomson

(1824–1907), who later matured into Lord Kelvin, realized that the work of

Carnot was not contradicting the ideas of Joule. One of his main contributions

remains a particular scale of absolute temperature. In his paper “On the dy-

namical theory of heat” appeared in 1851, Kelvin developed the point of view

that the mechanical action of heat could be interpreted by appealing to two

laws, later known as the ﬁrst and second laws. In this respect, Rudolf Clausius

(1822–1888), a contemporary of Joule and Kelvin, accomplished substantial

advancements. Clausius was the ﬁrst to introduce the words “internal energy”

and “entropy”, one of the most subtle notions of thermodynamics. Clausius

got deﬁnitively rid of the notion of caloric, reformulated Kelvin’s statement of

the second law, and tried to explain heat in terms of the behaviour of the indi-

vidual particles composing matter. It was the merit of Carnot, Joule, Kelvin,

and Clausius to thrust thermodynamics towards the level of an undisputed

scientiﬁc discipline. Another generation of scientists was needed to unify this

new formalism and to link it with other currents of science. One of them

was Ludwig Boltzmann (1844–1906) who put forward a decisive “mechanis-

tic” interpretation of heat transport; his major contribution was to link the

behaviour of the particles at the microscopic level to their consequences on

the macroscopic level. Another prominent scientist, Josiah Williard Gibbs

(1839–1903), deserves the credit to have converted thermodynamics into a

deductive science. In fact he recognized soon that thermodynamics of the

nineteenth century is a pure static science wherein time does not play any

1.2 Scope and Deﬁnitions 3

role. Among his main contributions, let us point out the theory of stability

based on the use of the properties of convex (or concave) functions, the po-

tential bearing his name, and the well-known Gibbs’ ensembles. Gibbs’ paper

“On the equilibrium of the heterogeneous substances” ranks among the most

decisive impacts in the developments of modern chemical thermodynamics.

Other leading scientists have contributed to the development of equilib-

rium thermodynamics as a well structured, universal, and undisputed science

since the pioneers laid down its ﬁrst steps. Although the list is far from being

exhaustive, let us mention the names of Caratheodory, Cauchy, Clapeyron,

Duhem, Einstein, Helmholtz, Maxwell, Nernst, and Planck.

1.2 Scope and Deﬁnitions

Equilibrium thermodynamics is a section of macroscopic physics whose orig-

inal objective is to describe the transformations of energy in all its forms. It

is a generalization of mechanics by introducing three new concepts:

1. The concept of state, i.e. an ensemble of quantities, called state variables,

whose knowledge allows us to identify any property of the system under

study. It is desirable that the state variables are independent and easily ac-

cessible to experiments. For example, a motionless ﬂuid may be described

by its mass m, volume V , and temperature T.

2. The notion of internal energy, complementing the notion of kinetic en-

ergy, which is of pure mechanical origin. Answering the question “what

is internal energy?” is a diﬃcult task. Internal energy is not a directly

measurable quantity: there exist no “energymeters”. For the moment, let

us be rather evasive and say that it is presumed to be some function of

the measurable properties of a system like mass, volume, and temperature.

Considering a macroscopic system as agglomerate of individual particles,

the internal energy can be viewed as the mean value of the sum of the

kinetic and interacting energies of the particles. The notion of internal en-

ergy is also related to these temperature and heat, which are absent from

the vocabulary of mechanics.

3. The notion of entropy. Like internal energy, it is a characteristic of the

system but we cannot measure it directly, we will merely have a way to

measure its changes. From a microscopic point of view, the notion of en-

tropy is related to disorder: the higher the entropy, the larger the disorder

inside the system. There are also connections between entropy and infor-

mation in the sense that entropy can be considered as a measure of our

lack of information on the state of the system. The link between entropy

and information is widely exploited into the so-called information theory.

Energy and entropy are obeying two major laws: the ﬁrst law stating that

the energy of the universe is a constant, and the second law stating that the

entropy of the universe never decreases.

4 1 Equilibrium Thermodynamics: A Review

At this stage, it is useful to recall some deﬁnitions. By system is understood

a portion of matter with a given mass, volume, and surface. An open system is

able to exchange matter and energy through its boundaries, a closed system

exchanges energy but not matter with the outside while an isolated system

does exchange neither energy nor matter with its surroundings. It is admitted

that the universe (the union of system and surroundings) acts as an isolated

system. In this chapter, we will deal essentially with homogeneous systems,

whose properties are independent of the position.

As mentioned earlier, the state of a system is deﬁned by an ensemble of

quantities, called state variables, characterizing the system. Considering a

system evolving between two equilibrium states, A and B, it is important to

realize that, by deﬁnition, the state variables will not depend on the partic-

ular way taken to go from A to B. The selection of the state variables is not

a trivial task, and both theoretical and experimental observations constitute

a suitable guide. It is to a certain extent arbitrary and non-unique, depend-

ing on the level of description, either microscopic or macroscopic, and the

degree of accuracy that is required. A delicate notion is that of equilibrium

state which turns out to be a state, which is time independent and generally

spatially homogeneous. It is associated with the absence of ﬂuxes of matter

and energy. On the contrary, a non-equilibrium state needs for its descrip-

tion time- and space-dependent state variables, because of exchanges of mass

and energy between the system and its surroundings. However, the above

deﬁnition of equilibrium is not complete; as shown in Sect. 1.3.3, equilibrium

of an isolated system is characterized by a maximum of entropy. Notice that

the concept of equilibrium is to some extent subjective; it is itself an idealiza-

tion and remains a little bit indeﬁnite because of the presence of ﬂuctuations

inherent to each equilibrium state. It depends also widely on the available

data and the degree of accuracy of our observations.

One distinguishes extensive and intensive state variables; extensive vari-

ables like mass, volume, and energy have values in a composite system equal

to the sum of the values in each subsystem; intensive variables as tempera-

ture or chemical potential take the same values everywhere in a system at

equilibrium. As a variable like temperature can only be rigorously deﬁned

at equilibrium, one may expect diﬃculties when dealing with situations

beyond equilibrium.

Classical thermodynamics is not ﬁrmly restricted to equilibrium states

but also includes the study of some classes of processes, namely those that

may be considered as a sequence of neighbouring equilibrium states. Such

processes are called quasi-static and are obtained by modifying the state

variables very slowly and by a small amount. A quasi-static process is either

reversible or irreversible. A reversible process 1 →2 →3 may be viewed as a

continuum sequence of equilibrium states and will take place inﬁnitesimally

slowly. When undergoing a reverse transformation 3 → 2 → 1, the state

variables take the same values as in the direct way and the exchanges of

matter and energy with the outside world are of opposite sign; needless to

1.3 The Fundamental Laws 5

say that reversible processes are pure idealizations. An irreversible process

is a non-reversible one. It takes place at ﬁnite velocity, may be mimicked

by a discrete series of equilibrium states and in a reverse transformation,

input of external energy from the outside is required to go back to its initial

state. Irreversible processes are generally associated with friction, shocks,

explosions, chemical reactions, viscous ﬂuid ﬂows, etc.

1.3 The Fundamental Laws

The ﬁrst law, also popularly known as the law of conservation of energy, was

not formulated ﬁrst but second after the second law, which was recognized

ﬁrst. Paradoxically, the zeroth law was formulated the latest, by Fowler during

the 1930s and quoted for the ﬁrst time in Fowler and Guggenheim’s book

published in 1939.

1.3.1 The Zeroth Law

It refers to the introduction of the idea of empirical temperature, which is one

of the most fundamental concepts of thermodynamics. When a system 1 is

put in contact with a system 2 but no net ﬂow of energy occurs, both systems

are said to be in thermal equilibrium. As sketched in Fig. 1.1a, we take two

systems 1 and 2, characterized by appropriate parameters, separated by an

adiabatic wall, but in contact (a thermal contact) with the system 3 through

a diathermal wall, which allows for energy transfer in opposition with an

adiabatic wall. If the systems 1 and 2 are put in contact (see Fig. 1.1b), they

will change the values of their parameters in such a way that they reach a

state of thermal equilibrium, in which there is no net heat transfer between

them.

Fig. 1.1 Steps for introducing the empirical temperature concept

6 1 Equilibrium Thermodynamics: A Review

The zeroth law of thermodynamics states that if the systems 1 and 2 are

separately in thermal equilibrium with 3, then 1 and 2 are in thermal equilib-

rium with one another. The property of transitivity of thermal equilibrium

allows one to classify the space of thermodynamic states in classes of equiv-

alence, each of which constituted by states in mutual thermal equilibrium.

Every class may be assigned a label, called empirical temperature, and the

mathematical relation describing a class in terms of its state variables and

the empirical temperature is known as the thermal equation of state of the

system. For one mole of a simple ﬂuid this equation has the general form

φ(p, V, θ) = 0 where p is the pressure, V the volume, and θ the empirical

temperature.

1.3.2 The First Law or Energy Balance

The ﬁrst law introduces the notion of energy, which emerges as a unifying

concept, and the notion of heat, related to the transfer of energy. Here, we

examine the formulation of the ﬁrst law for closed systems.

Consider ﬁrst a system enclosed by a thermally isolated (adiabatic), imper-

meable wall, so that the sole interaction with the external world will appear

under the form of a mechanical work W, for instance by expansion of its

volume or by stirring. Referring to the famous experience of Joule, the work

can be measured by the decrease in potential energy of a slowly falling weight

and is given by W = mgh, where h is the displacement and g the acceler-

ation of gravity. During the evolution of the system between the two given

equilibrium states A and B, it is checked experimentally that the work W is

determined exclusively by the initial and the ﬁnal states A and B, indepen-

dently of the transformation paths. This observation allows us to identify W

with the diﬀerence ∆U = U(B) − U(A) of a state variable U which will be

given the name of internal energy

W = ∆U. (1.1)

The above result provides a mean to measure the internal energy of a

system, whatever be its nature. Assume now that we remove the adiabatic

wall enclosing the system, which again proceeds from state A to state B.

When this is accomplished, it is observed that in general W ,= ∆U, and

calling Q the diﬀerence between these two quantities, one obtains

∆U −W = Q, (1.2)

where Q is referred to as the heat exchanged between the system and its sur-

roundings. Expression (1.2) is the ﬁrst law of thermodynamics and is usually

written under the more familiar form

∆U = Q+W, (1.3a)

1.3 The Fundamental Laws 7

or, in terms of diﬀerentials,

dU = ¯ dQ+ ¯ dW, (1.3b)

where the stroke through the symbol “d” means that ¯ dQ and ¯ dW are inexact

diﬀerentials, i.e. that they depend on the path and not only on the initial and

ﬁnal states. From now on, we adopt the sign convention that Q > 0, W > 0

when heat and work are supplied to the system, Q < 0, W < 0 when heat

and work are delivered by the system. Some authors use other conventions

resulting in a minus sign in front of ¯ dW.

It is important to stress that the domain of applicability of the ﬁrst law

is not limited to reversible processes between equilibrium states. The ﬁrst

law remains valid whatever the nature of the process, either reversible or

irreversible and the status of the states A and B, either equilibrium or non-

equilibrium. Designating by E = U +K+E

pot

the total energy of the system

(i.e. the sum of the internal U, kinetic K, and potential energy E

pot

), (1.3b)

will be cast in the more general form

dE = ¯ dQ+ ¯ dW. (1.4)

At this point, it should be observed that with respect to the law of energy

∆K = W as known in mechanics, we have introduced two new notions:

internal energy U and heat Q. The internal energy can be modiﬁed either

by heating the body or by acting mechanically, for instance by expansion or

compression, or by coupling both mechanisms. The quantity U consists of

a stored energy in the body while Q and W represent two diﬀerent means

to transfer energy through its boundaries. The internal energy U is a state

function whose variation is completely determined by the knowledge of the

initial and ﬁnal states of the process; in contrast, Q and W are not state

functions as they depend on the particular path followed by the process. It

would therefore be incorrect to speak about the heat or the work of a system.

The diﬀerence between heat and work is that the second is associated with

a change of the boundaries of the system or of the ﬁeld acting on it, like a

membrane deformation or a piston displacement. Microscopically, mechanical

work is related to coherent correlated motions of the particles while heat

represents that part of motion, which is uncorrelated, say incoherent.

In equilibrium thermodynamics, the processes are reversible from which

follows that the energy balance equation (1.4) will take the form:

dU = ¯ dQ

rev

−p dV, (1.5)

wherein use is made of the classical result that the reversible work per-

formed by a piston that compress a gas of volume V and pressure p trapped

in a cylinder is given by ¯ dW

rev

= −p dV (see Problem 1.1). In engineer-

ing applications, it is customary to work with the enthalpy H deﬁned by

H = U +pV . In terms of H, expression (1.5) of the ﬁrst law reads as

dH = ¯ dQ

rev

+V dp. (1.6)

8 1 Equilibrium Thermodynamics: A Review

For an isolated system, one has simply

dU = 0 (1.7)

expressing that its energy remains constant.

Note that, when applied to open systems with n diﬀerent constituents,

(1.5) will contain an additional contribution due to the exchange of matter

with the environment and takes the form (Prigogine 1947)

dU = d

i

Q−p dV +

n

k=1

h

k

d

e

m

k

; (1.8)

note that d

i

Q is not the total amount of heat but only that portion associ-

ated to the variations of the thermomechanical properties, T and p, and the

last term in (1.8), which is the extra contribution caused by the exchange

of matter d

e

m

k

with the surroundings, depends on the speciﬁc enthalpy

h

k

= H/m

k

of the various constituents.

1.3.3 The Second Law

The ﬁrst law does not establish any preferred direction for the evolution of

the system. For instance, it does not forbid that heat could pass sponta-

neously from a body of lower temperature to a body of higher temperature,

nor the possibility to convert completely heat into work or that the huge

energy contained in oceans can be transformed in available work to propel

a boat without consuming fuel. More generally, the ﬁrst law establishes the

equivalence between heat and work but is silent about the restrictions on the

transformation of one into the other. The role of the second law of thermo-

dynamics is to place such limitations and to reﬂect the property that natural

processes evolve spontaneously in one direction only. The ﬁrst formulations of

the second law were proposed by Clausius (1850, 1851) and Kelvin (1851) and

were stated in terms of the impossibility of some processes to be performed.

Clausius’ statement of the second law is enunciated as follows: No process is

possible whose sole eﬀect is to transfer heat from a cold body to a hot body.

Kelvin’s statement considers another facet: it is impossible to construct an

engine which can take heat from a single reservoir, and convert it entirely to

work in a cyclic process. In this book we will examine in detail, the formu-

lations of the second law out of equilibrium. Here, we shall concentrate on

some elements that are essential to a good understanding of the forthcoming

chapters. We will split the presentation of the second law in two parts. In

the ﬁrst one, we are going to build-up a formal deﬁnition of a new quantity,

the entropy – so named by Clausius from the Greek words en (in) and trope

(turning) for representing “capacity of change or transformation” – which

is as fundamental and universal (for equilibrium systems) as the notion of

1.3 The Fundamental Laws 9

energy. In the second part, which constitutes truly the essence of the second

law, we shall enounce the principle of entropy increase during an irreversible

process.

1.3.3.1 The Concept of Entropy

Consider a homogeneous system of constant mass undergoing a reversible

transformation between two equilibrium states A and B. The quantity of

heat

_

B

A

¯ dQ

rev

depends on the path followed between states A and B (in

mathematical terms, it is an imperfect diﬀerential) and therefore cannot be

selected as a state variable. However, experimental observations have indi-

cated that by dividing ¯ dQ

rev

by a uniform and continuous function T(θ) of

an empirical temperature θ, one obtains an integral which is independent of

the path and may therefore be identiﬁed with a state function, called entropy

and denoted S

_

B

A

¯ dQ

rev

T(θ)

= ∆S = S

B

−S

A

. (1.9)

Since in reversible processes, quantities of heat are additive, entropy is also

additive and is thus an extensive quantity. A function like T(θ) which trans-

forms an imperfect diﬀerential into a perfect one is called an integrating

factor. The empirical temperature is that indicated by a mercury or an alco-

hol thermometer or a thermocouple and its value depends of course on the

nature of the thermometer; the same remark is true for the entropy, as it

depends on T(θ). It was the great merit of Kelvin to propose a temperature

scale for T, the absolute temperature, independently of any thermodynamic

system (see Box 1.1). In diﬀerential terms, (1.9) takes the form

dS =

¯ dQ

rev

T

. (1.10)

This is a very important result as it introduces two new concepts, absolute

temperature and entropy. The latter can be viewed as the quantity of heat

exchanged by the system during a reversible process taking place at the

equilibrium temperature T. Note that only diﬀerences in entropy can be

measured. Given two equilibrium states A and B, it is always possible to

determine their entropy diﬀerence regardless of whether the process between

A and B is reversible or irreversible. Indeed, it suﬃces to select or imagine

a reversible path joining these initial and ﬁnal equilibrium states. The ques-

tion is how to realize a reversible heat transfer. Practically, the driving force

for heat transfer is a temperature diﬀerence and for reversible transfer, we

need only imagine that this temperature diﬀerence is inﬁnitesimally small so

that ¯ dQ

rev

= lim

∆T→0

¯ dQ. Nevertheless, when the process takes place between

non-equilibrium states, the problem of the deﬁnition of entropy is open, and

actually not yet deﬁnitively solved.

10 1 Equilibrium Thermodynamics: A Review

Box 1.1 Absolute Temperature

Heat engines take heat from some hot reservoir, deliver heat to some cold

reservoir, and perform an amount of work, i.e. they partially transform heat

into work. Consider a Carnot’s reversible engine (see Fig. 1.2a) operating

between a single hot reservoir at the unknown empirical temperature θ

1

and

a single cold reservoir at temperature θ

2

. The Carnot cycle is accomplished

in four steps consisting in two isothermal and two adiabatic transformations

(Fig. 1.2b).

During the ﬁrst isothermal process, the Carnot’s engine absorbs an

amount of heat Q

1

at temperature θ

1

. In the second step, the system un-

dergoes an adiabatic expansion decreasing the temperature from θ

1

to θ

2

.

Afterwards, the system goes through an isothermal compression at tem-

perature θ

2

(step 3) and ﬁnally (step 4), an adiabatic compression which

brings the system back to its initial state. After one cycle, the engine has

performed a quantity of work W but its total variation of entropy is zero

∆S

engine

=

[Q

1

[

T(θ

1

)

−

[Q

2

[

T(θ

2

)

= 0. (1.1.1)

Selecting the reference temperature as T(θ

2

) = 273.16, the triple point

temperature of water, it follows from (1.1.1)

T(θ

1

) = 273.16

[Q

1

[

[Q

2

[

. (1.1.2)

The ratio [Q

1

[ / [Q

2

[ is universal in the sense that it is independent of the

working substance. Therefore, Carnot cycles oﬀer the opportunity to re-

duce temperature measurements to measurements of quantities of heat and

to deﬁne an absolute scale of positive temperatures, independently of the

measurement of temperature on any empirical temperature scale, which

depends on thermometric substance.

Fig. 1.2 (a) Heat engine and (b) Carnot diagram

1.3 The Fundamental Laws 11

The eﬃciency of a heat engine, in particular that of Carnot, is deﬁned by

the ratio of the work produced to the heat supplied

η =

W

Q

1

(1.11)

for a cycle one has, in virtue of the ﬁrst law, W = Q

1

−Q

2

, so that

η = 1 −

Q

2

Q

1

. (1.12)

Finally, making use of (1.1.2), it is found that the eﬃciency of a reversible

cycle is

η = 1 −

T

2

T

1

. (1.13)

As it will be seen, this is the maximum value for the eﬃciency of any heat en-

gine working between the selected heat reservoirs. More considerations about

the eﬃciency of reversible and irreversible cycles are developed in Chap. 5.

1.3.3.2 The Principle of Increase of Entropy

The second law was formulated by Clausius (1865) for isolated systems in

terms of the change of the entropy in the form

∆S ≥ 0. (1.14)

To illustrate the principle of entropy increase, imagine an arbitrary number of

subsystems, for instance three diﬀerent gases A, B, and C at equilibrium, en-

closed in a common isolated container and separated each other by adiabatic

and rigid walls (Fig. 1.3). Let S

ini

be the entropy in this initial conﬁguration.

Remove then the internal wall separating A and B which are diﬀusing into

each other until a new state of equilibrium characterized by an entropy S

int

,

corresponding to the intermediate conﬁguration, which is larger than S

ini

is

reached. By eliminating ﬁnally the last internal constraint between A∪B and

C, and after the ﬁnal state of equilibrium, corresponding to complete mixing,

is reached, it is noted that entropy S

ﬁn

is still increased: S

ﬁn

> S

int

> S

ini

.

Figure 1.3 reﬂects also that disorder is increased by passing from the initial

Fig. 1.3 Increase of entropy after removal of internal constraints

12 1 Equilibrium Thermodynamics: A Review

to the ﬁnal conﬁguration, which suggests the use of entropy as a measure of

disorder: larger the disorder larger the entropy (Bridgman 1941).

It is therefore concluded that entropy is increased as internal constraints

are removed and that entropy reaches a maximum in the ﬁnal state of equi-

librium, i.e. the state of maximum “disorder”. In other terms, in isolated

systems, one has

∆S = S

ﬁn

−S

in

≥ 0 (isolated system). (1.15)

Thus, entropy is continuously increasing when irreversible processes take

place until it reaches a state of maximum value, the equilibrium state, which

in mathematical terms is characterized by dS = 0, d

2

S < 0. This statement

constitutes the celebrated principle of entropy increase and is often referred

to as the Second Law of thermodynamics. It follows that a decrease in entropy

dS < 0 corresponds to an impossible process. Another consequence is that

the entropy of an isolated system remains constant when reversible processes

occur in it.

An illustration of the entropy increase principle is found in Box 1.2. When

the system is not isolated, as in the case of closed and open systems, the

entropy change in the system consists in two parts: d

e

S due to exchanges of

energy and matter with the outside, which may be positive or negative, and

d

i

S due to internal irreversible processes

dS = d

e

S + d

i

S. (1.16)

The second law asserts that the entropy production d

i

S can only be greater

than or equal to zero

d

i

S ≥ 0 (closed and open systems), (1.17)

the equality sign referring to reversible or equilibrium situations. Expres-

sion (1.17) is the statement of the second law in its more general form. In

the particular case of isolated systems, there is no exchange of energy and

matter so that d

e

S = 0 and one recovers (1.15) of the second law, namely

dS = d

i

S ≥ 0. For closed systems, for which d

e

S = ¯ dQ/T, one has

dS ≥ ¯ dQ/T (closed system). (1.18)

In the particular case of a cyclic process for which dS=0, one has ¯ dQ/T≤0,

which is usually identiﬁed as the Clausius’ inequality.

Box 1.2 Entropy Increase

Consider two diﬀerent gases A and B at equilibrium, enclosed in a com-

mon isolated container and separated each other by an adiabatic and ﬁxed

wall (Fig. 1.4). Both gases are characterized by their internal energy U and

volume V .

1.3 The Fundamental Laws 13

In the initial conﬁguration, entropy S

(i)

is a function of the initial values

of internal energy U

(i)

A

and volume V

(i)

A

corresponding to subsystem A,

and similarly of U

(i)

B

and V

(i)

B

for subsystem B, in such a way that S

(i)

=

S

A

(U

(i)

A

, V

(i)

A

) + S

B

(U

(i)

B

, V

(i)

B

). If the adiabatic and ﬁxed wall separating

both subsystems A and B is replaced by a diathermal and movable wall, a

new conﬁguration is attained whose entropy S

(f)

may be expressed as S

(f)

=

S

A

(U

(f)

A

, V

(f)

A

) + S

B

(U

(f)

B

, V

(f)

B

); superscript (f) denotes the ﬁnal values of

energy and volume submitted to the closure relations U

(i)

A

+ U

(i)

B

= U

(f)

A

+

U

(f)

B

= U

total

and V

(i)

A

+ V

(i)

B

= V

(f)

A

+ V

(f)

B

= V

total

reﬂecting conservation

of these quantities for the composite system A+B. The removal of internal

constraints that prevent the exchange of internal energy and volume leads

to the establishment of a new equilibrium state of entropy S

(f)

> S

(i)

.

The values taken by the (extensive) variables, in the absence of internal

constraints, in this case U

(f)

A

, V

(f)

A

and U

(f)

B

, V

(f)

B

, are those that maximize

the entropy over the manifold of equilibrium states (Callen 1985).

In Fig. 1.4 is represented S

(f)

/S

(i)

in terms of x ≡ U

A

/U

total

and y ≡

V

A

/V

total

using an ideal gas model; the ﬁnal values of x and y are those

corresponding to the maximum of S

(f)

/S

(i)

. The arbitrary curve drawn on

the surface between the initial “i” and ﬁnal “f” states stands for an idealized

process deﬁned as a succession of equilibrium states, quite distinct from a

real physical process formed by a temporal succession of equilibrium and

non-equilibrium states.

Fig. 1.4 Illustration of the entropy increase principle in the case of two gases initially

separated by an adiabatic and ﬁxed wall

14 1 Equilibrium Thermodynamics: A Review

1.3.4 The Third Law

The roots of this law appear in the study of thermodynamic quantities as

the absolute temperature tends to zero. In 1909, Nernst formulated his heat

theorem, later known as the third law of thermodynamics, to better under-

stand the nature of chemical equilibrium. Nernst’s formulation was that the

entropy change in any isothermal process approaches zero as the temperature

at which the process occurs approaches zero, i.e.

(∆S)

T→0

→0. (1.19)

This statement is suﬃcient for any thermodynamic development, but some-

times the stronger Planck’s statement (S → 0 as T → 0) is preferred. Since

the third law is more of quantum statistical essence, it is not of the same

nature as the other laws and no further reference will be made to it in this

book.

1.4 Gibbs’ Equation

Let us now gather the results obtained for the ﬁrst and second laws. Consider

a reversible transformation, taking place in a closed system, for which the ﬁrst

law takes the form

dU = ¯ dQ

rev

−p dV, (1.20)

and combine it with the deﬁnition of entropy ¯ dQ

rev

= T dS, resulting in

dU = T dS −p dV. (1.21)

Expression (1.21) is known as Gibbs’ equation; it is, however, not complete

when there are matter exchanges as in open systems, or variations in compo-

sition as in chemical reactions. To calculate the reversible work corresponding

to a chemical reaction involving n species, it is necessary to devise a reversible

process of mixing. This is achieved thanks to van’t Hoﬀ’s box (Kestin 1968),

accordingly the reversible chemical work is given by

¯ dW

ch

rev

=

n

k=1

¯ µ

k

dm

k

, (1.22)

where ¯ µ

k

is deﬁned as the chemical potential of substance k. The properties of

the chemical potential will be explicitly examined below. With this additional

term, one is led to the generalized Gibbs’ equations

dU = T dS −p dV +

n

k=1

¯ µ

k

dm

k

, (1.23a)

1.4 Gibbs’ Equation 15

or equivalently,

dS = T

−1

dU +pT

−1

dV −

n

k=1

T

−1

¯ µ

k

dm

k

. (1.23b)

As discussed in the forthcoming sections, the Gibbs’ equation plays a fun-

damental role in equilibrium thermodynamics. We should also mention that

Gibbs’ equation is one of the pillars of the Classical Theory of Irreversible

Processes, as shown in Chap. 2. Let us now examine the main consequences

of Gibbs’ equation.

1.4.1 Fundamental Relations and State Equations

It follows directly from Gibbs’ equation (1.23a) that

U = U(S, V, m

1

, m

2

, . . . , m

n

), (1.24)

or, solving with respect to S,

S = S(U, V, m

1

, m

2

, . . . , m

n

). (1.25)

Relations like (1.24) or (1.25) expressing that U or S are single-valued func-

tions of extensive state variables are called fundamental relations because

they contain all thermodynamic information about the system. When U (re-

spectively, S) is expressed as a function of the variables, we are speaking of

the “energy representation” (respectively, “entropy representation”).

Another consequence of Gibbs’ equation (1.23a) is that the intensive vari-

ables, represented by temperature, pressure and chemical potentials, can be

deﬁned as partial derivatives of U:

T =

_

∂U

∂S

_

V,{m

k

}

(a), p = −

_

∂U

∂V

_

S,{m

k

}

(b), ¯ µ

k

=

_

∂U

∂m

k

_

V,S,{m

i=k

}

(c),

(1.26)

where ¦m

k

¦ stands for all m

k

constant. Since U is a function of S, V , m

k

,

the same remains true for T, p, and µ

k

so that

T = T(S, V, m

1

, m

2

, . . . , m

n

), (1.27a)

p = p(S, V, m

1

, m

2

, . . . , m

n

), (1.27b)

¯ µ

k

= ¯ µ

k

(S, V, m

1

, m

2

, . . . , m

n

). (1.27c)

Such relationships between intensive and extensive variables are called state

equations. Elimination of S between (1.27a) and (1.27b) leads to the ther-

mal equation of state p = p(T, V, m

1

, m

2

, . . . , m

n

); similarly by combining

16 1 Equilibrium Thermodynamics: A Review

(1.24) and (1.27a), one obtains the so-called caloric equation U = U(T, V, m

1

,

m

2

, . . . , m

n

). The knowledge of one single state equation is not suﬃcient to

describe the state of a system, which requires the knowledge of all the equa-

tions of state. For instance in the case of a monatomic perfect gas, pV = NRT

does not constitute the complete knowledge of the system but must be com-

plemented by U =

3

2

NRT, R being the gas constant and N the mole number.

1.4.2 Euler’s Relation

The extensive property of U implies that, from the mathematical point of

view, it is a ﬁrst-order homogeneous function of the extensive variables:

U(λS, λV, λm

1

, . . . , λm

n

) = λU(S, V, m

1

, . . . , m

n

), (1.28)

where λ is an arbitrary scalar. Diﬀerentiation of the fundamental relation

(1.28) with respect to λ and setting λ = 1, leads to

_

∂U

∂S

_

V,{m

k

}

S +

_

∂U

∂V

_

S,{m

k

}

V +

k

_

∂U

∂m

k

_

V,S,{m

j=k

}

m

k

= U, (1.29)

and, after making use of (1.26), one obtains Euler’s relation

U = TS −pV +

k

¯ µ

k

m

k

. (1.30)

1.4.3 Gibbs–Duhem’s Relation

A diﬀerential equation among the intensive variables can be derived directly

from Euler’s relation. Indeed, after diﬀerentiating (1.30), it is found that

dU = T dS −p dV +

n

k=1

¯ µ

k

dm

k

+S dT −V dp +

n

k=1

m

k

d¯ µ

k

, (1.31)

which, after using Gibbs’ equation (1.23a), yields Gibbs–Duhem’s relation

S dT −V dp +

n

k=1

m

k

d¯ µ

k

= 0. (1.32)

It follows that the n + 2 intensive variables are not independent but related

through the Gibbs–Duhem’s relation. For a n-component mixture, the num-

ber of independent intensive state variables, called thermodynamic degrees

of freedom, is equal to n + 1: for instance, the n − 1 chemical potentials

plus temperature and pressure. In the case of a one-component ﬂuid, the

thermodynamic description of the system requires the knowledge of two in-

dependent intensive quantities, generally selected as the temperature T and

the pressure p.

1.4 Gibbs’ Equation 17

1.4.4 Some Deﬁnitions

In view of further developments, it is useful to introduce the following deﬁn-

itions of well-known experimental quantities:

• Coeﬃcient of thermal expansion:

α =

1

V

_

∂V

∂T

_

p,{m

k

}

. (1.33)

• Isothermal compressibility:

κ

T

= −

1

V

_

∂V

∂p

_

T,{m

k

}

. (1.34)

• Heat capacity at constant volume:

c

V

=

_

∂U

∂T

_

V,{m

k

}

=

_

¯ dQ

rev

dT

_

V,{m

k

}

. (1.35)

• Heat capacity at constant pressure:

c

p

=

_

∂H

∂T

_

p,{m

k

}

=

_

¯ dQ

rev

dT

_

p,{m

k

}

. (1.36)

Other partial derivatives may be introduced but generally, they do not have a

speciﬁc practical usefulness. Relations between these partial derivatives may

be derived by equating mixed second-order partial derivatives of U and S.

Such expressions have been identiﬁed as Maxwell’s relations.

As a last remark, let us mention that the results established so far in

homogeneous systems of total mass m and volume V are still valid when

referred per unit mass and unit volume. Analogous to (1.24), the fundamental

relation per unit mass is

u = u(s, v, . . . , c

k

, . . .) (1.37)

with u = U/m, s = S/m, v = V/m, c

k

= m

k

/m, and

k

c

k

= 1. After

diﬀerentiation, (1.37) reads as

du = T ds −p dv +

n−1

k=1

(¯ µ

k

− ¯ µ

n

)dc

k

(1.38)

with T = (∂u/∂s)

v,{c

k

}

, p = −(∂u/∂v)

s,{c

k

}

, ¯ µ

k

= (∂u/∂c

k

)

s,v,{c

j=k

}

. Simi-

larly, the Euler and Gibbs–Duhem’s relations (1.30) and (1.32) take the form

u = Ts −pv +

n

k=1

¯ µ

k

c

k

, S dT −v dp +

n

k=1

c

k

d¯ µ

k

= 0. (1.39)

18 1 Equilibrium Thermodynamics: A Review

1.4.5 The Basic Problem of Equilibrium

Thermodynamics

To maintain a system in an equilibrium state, one needs the presence of

constraints; if some of them are removed, the system will move towards a new

equilibrium state. The basic problem is to determine the ﬁnal equilibrium

state when the initial equilibrium state and the nature of the constraints

are speciﬁed. As illustration, we have considered in Box 1.3 the problem

of thermo-diﬀusion. The system consists of two gases ﬁlling two containers

separated by a rigid, impermeable and adiabatic wall: the whole system is

isolated. If we now replace the original wall by a semi-permeable, diathermal

one, there will be heat exchange coupled with a ﬂow of matter between the

two subsystems until a new state of equilibrium is reached; the problem is

the calculation of the state parameters in the ﬁnal equilibrium state.

Box 1.3 Thermodiﬀusion

Let us suppose that an isolated system consists of two separated containers

I and II, each of ﬁxed volume, and separated by an impermeable, rigid

and adiabatic wall (see Fig. 1.5). Container I is ﬁlled with a gas A and

container II with a mixture of two non-reacting gases A and B. Substitute

now the original wall by a diathermal, non-deformable but semi-permeable

membrane, permeable to substance A. The latter will diﬀuse through the

membrane until the system comes to a new equilibrium, of which we want

to know the properties. The volumes of each container and the mass of

substance B are ﬁxed:

V

I

= constant, V

II

= constant, m

B

II

= constant, (1.3.1)

but the energies in both containers as well as the mass of substance A are

free to change, subject to the constraints

U

I

+U

II

= constant, m

A

I

+m

A

II

= constant. (1.3.2)

In virtue of the second law, the values of U

I

, U

II

, m

A

I

, m

A

II

in the new equi-

librium state are such as to maximize the entropy, i.e. dS = 0 and, from

the additivity of the entropy in the two subsystems

Fig. 1.5 Equilibrium conditions for thermodiﬀusion

1.5 Legendre Transformations and Thermodynamic Potentials 19

dS = dS

I

+ dS

II

= 0. (1.3.3)

Making use of the Gibbs’ relation (1.23b) and the constraints (1.3.1) and

(1.3.2), one may write

dS =

∂S

I

∂U

I

dU

I

+

∂S

I

∂m

A

I

dm

A

I

+

∂S

II

∂U

II

dU

II

+

∂S

II

∂m

A

II

dm

A

II

=

_

1

T

I

−

1

T

II

_

dU

I

−

_

¯ µ

A

I

T

I

−

¯ µ

A

II

T

II

_

dm

A

I

= 0.

(1.3.4)

Since this relation must be satisﬁed for arbitrary variations of U

I

and m

A

I

,

one ﬁnds that the equilibrium conditions are that

T

I

= T

II

, ¯ µ

A

I

= ¯ µ

A

II

. (1.3.5)

The new equilibrium state, which corresponds to absence of ﬂow of sub-

stance A, is thus characterized by the equality of temperatures and chemical

potentials in the two containers.

In absence of mass transfer, only heat transport will take place. During

the irreversible process between the initial and ﬁnal equilibrium states, the

only admissible exchanges are those for which

dS =

_

1

T

I

−

1

T

II

_

¯ dQ

I

> 0, (1.3.6)

where use has been made of the ﬁrst law dU

I

= ¯ dQ

I

. If T

I

> T

II

, one has

¯ dQ

I

< 0 while for T

I

< T

II

, ¯ dQ

I

> 0 meaning that heat will spontaneously

ﬂow from the hot to the cold container. The formal restatement of this

item is the Clausius’ formulation of the second law: “no process is possible

in which the sole eﬀect is transfer of heat from a cold to a hot body”.

Under isothermal conditions (T

I

= T

II

), the second law imposes that

dS =

1

T

I

_

¯ µ

A

II

− ¯ µ

A

I

_

dm

A

I

> 0 (1.3.7)

from which is concluded that matter ﬂows spontaneously from regions of

high to low chemical potential.

1.5 Legendre Transformations and Thermodynamic

Potentials

Although the fundamental relations (1.24) and (1.25) that are expressed in

terms of extensive variables are among the most important, they are not

the most useful. Indeed, in practical situations, the intensive variables, like

20 1 Equilibrium Thermodynamics: A Review

Box 1.4 Legendre Transformations

The problem to be solved is the following: given a fundamental relation of

the extensive variables A

1

, A

2

, . . . , A

n

,

Y = Y (A

1

, A

2

, . . . , A

k

, A

k+1

, . . . , A

n

) (1.4.1)

ﬁnd a new function for which the derivatives

P

i

=

∂Y

∂A

i

(i = 1, . . . , k ≤ n) (1.4.2)

will be considered as the independent variables instead of A

1

, . . . , A

k

. The

solution is given by

Y [P

1

, . . . , P

k

] = Y −

k

i=1

P

i

A

i

. (1.4.3)

Indeed, taking the inﬁnitesimal variation of (1.4.3) results in

dY [P

1

, . . . , P

k

] = −

k

1

A

i

dP

i

+

n

k+1

P

i

dA

i

, (1.4.4)

which indicates clearly that Y [P

1

, . . . , P

k

] is a function of the indepen-

dent variables P

1

, . . . , P

k

, A

k+1

, . . . , A

n

. With Callen (1985), we have used

the notation Y [P

1

, . . . , P

k

] to denote the partial Legendre transformation

with respect to A

1

, . . . , A

k

. The function Y [P

1

, . . . , P

k

] is referred to as a

Legendre transformation.

temperature and pressure, are more easily measurable and controllable. In

contrast, there is no instrument to measure directly entropy and internal

energy. This observation has motivated a reformulation of the theory, in which

the central role is played by the intensive rather than the extensive quantities.

Mathematically, this is easily achieved thanks to the introduction of Legendre

transformations, whose mathematical basis is summarized in Box 1.4.

1.5.1 Thermodynamic Potentials

The application of the preceding general considerations to thermodynamics

is straightforward: the derivatives P

1

, P

2

, . . . will be identiﬁed with the inten-

sive variables T, −p, µ

k

and the several Legendre transformations are known

as the thermodynamic potentials. Starting from the fundamental relation,

U = U(S, V, m

k

), replace the entropy S by ∂U/∂S ≡ T as independent vari-

able, the corresponding Legendre transform is, according to (1.4.3),

1.5 Legendre Transformations and Thermodynamic Potentials 21

U[T] ≡ F = U −

_

∂U

∂S

_

V,{m

k

}

S = U −TS, (1.40)

which is known as Helmholtz’s free energy. Replacing the volume V by

∂U/∂V ≡ −p, one deﬁnes the enthalpy H as

U[p] ≡ H = U −

_

∂U

∂V

_

S,{m

k

}

V = U +pV. (1.41)

The Legendre transform which replaces simultaneously S by T and V by −p

is the so-called Gibbs’ free energy G given by

U[T, p] ≡ G = U −TS +pV =

n

k=1

m

k

¯ µ

k

. (1.42)

The last equality has been derived by taking account of Euler’s relation (1.30).

Note that the complete Legendre transform

U[T, p, µ

1

, . . . , µ

r

] = U −TS +pV −

n

k=1

¯ µ

k

m

k

= 0 (1.43)

is identically equal to zero in virtue of Euler’s relation and this explains why

only three thermodynamic potentials can be deﬁned from U. The fundamen-

tal relations of F, H, and G read in diﬀerential form:

dF = −S dT −p dV +

n

k=1

¯ µ

k

dm

k

, (1.44a)

dH = T dS +V dp +

n

k=1

¯ µ

k

dm

k

, (1.44b)

dG = −S dT +V dp +

n

k=1

¯ µ

k

dm

k

. (1.44c)

Another set of Legendre transforms can be obtained by operating on the en-

tropy S = S(U, V, m

1

, . . . , m

n

), and are called the Massieu–Planck functions,

particularly useful in statistical mechanics.

1.5.2 Thermodynamic Potentials and Extremum

Principles

We have seen that the entropy of an isolated system increases until it attains

a maximum value: the equilibrium state. Since an isolated system does not

exchange heat, work, and matter with the surroundings, it will therefore be

22 1 Equilibrium Thermodynamics: A Review

characterized by constant values of energy U, volume V , and mass m. In

short, for a constant mass, the second law can be written as

dS ≥ 0 at U and V constant. (1.45)

Because of the invertible roles of entropy and energy, it is equivalent to for-

mulate the second principle in terms of U rather than S.

1.5.2.1 Minimum Energy Principle

Let us show that the second law implies that, in absence of any internal

constraint, the energy U evolves to a minimum at S and V ﬁxed:

dU ≤ 0 at S and V constant. (1.46)

We will prove that if energy is not a minimum, entropy is not a maximum

in equilibrium. Suppose that the system is in equilibrium but that its in-

ternal energy has not the smallest value possible compatible with a given

value of the entropy. We then withdraw energy in the form of work, keeping

the entropy constant, and return this energy in the form of heat. Doing so,

the system is restored to its original energy but with an increased value of the

entropy, which is inconsistent with the principle that the equilibrium state is

that of maximum entropy.

Since in most practical situations, systems are not isolated, but closed and

then subject to constant temperature or (and) constant pressure, it is appro-

priate to reformulate the second principle by incorporating these constraints.

The evolution towards equilibrium is no longer governed by the entropy or

the energy but by the thermodynamic potentials.

1.5.2.2 Minimum Helmholtz’s Free Energy Principle

For closed systems maintained at constant temperature and volume, the

leading potential is Helmholtz’s free energy F. In virtue of the deﬁnition

of F(= U −TS), one has, at constant temperature,

dF = dU −T dS, (1.47)

and, making use of the ﬁrst law and the decomposition dS = d

e

S + d

i

S,

dF =

¯

dQ−p dV −T d

e

S −T d

i

S. (1.48)

In closed systems d

e

S = ¯ dQ/T and, if V is maintained constant, the change

of F is

dF = −T d

i

S ≤ 0. (1.49)

1.5 Legendre Transformations and Thermodynamic Potentials 23

It follows that closed systems at ﬁxed values of the temperature and the

volume, are driven towards an equilibrium state wherein the Helmholtz’s

free energy is minimum. Summarizing, at equilibrium, the only admissible

processes are those satisfying

dF ≤ 0 at T and V constant. (1.50)

1.5.2.3 Minimum Enthalpy Principle

Similarly, the enthalpy H = U +pV can also be associated with a minimum

principle. At constant pressure, one has

dH = dU +p dV = ¯ dQ, (1.51)

but for closed systems, ¯ dQ = T d

e

S = T(dS − d

i

S), whence, at ﬁxed values

of p and S,

dH = −T d

i

S ≤ 0, (1.52)

as a direct consequence of the second law. Therefore, at ﬁxed entropy and

pressure, the system evolves towards an equilibrium state characterized by a

minimum enthalpy, i.e.

dH ≤ 0 at S and p constant. (1.53)

1.5.2.4 Minimum Gibbs’ Free Energy Principle

Similar considerations are applicable to closed systems in which both tem-

perature and pressure are maintained constant but now the central quantity

is Gibbs’ free energy G = U −TS +pV . From the deﬁnition of G, one has at

T and p ﬁxed,

dG = dU −T dS +p dV = ¯ dQ−T(d

e

S + d

i

S) = −T d

i

S ≤ 0, (1.54)

wherein use has been made of d

e

S = ¯ dQ/T. This result tells us that a closed

system, subject to the constraints T and p constant, evolves towards an equi-

librium state where Gibbs’ free energy is a minimum, i.e.

dG ≤ 0 at T and p constant. (1.55)

The above criterion plays a dominant role in chemistry because chemical

reactions are usually carried out under constant temperature and pressure

conditions.

24 1 Equilibrium Thermodynamics: A Review

It is left as an exercise (Problem 1.7) to show that the (maximum) work

delivered in a reversible process at constant temperature is equal to the de-

crease in the Helmholtz’s free energy:

¯ dW

rev

= −dF. (1.56)

This is the reason why engineers call frequently F the available work at con-

stant temperature. Similarly, enthalpy and Gibbs’ free energy are measures

of the maximum available work at constant p, and at constant T and p,

respectively.

As a general rule, it is interesting to point out that the Legendre trans-

formations of energy are a minimum for constant values of the transformed

intensive variables.

1.6 Stability of Equilibrium States

Even in equilibrium, the state variables do not keep rigorous ﬁxed values

because of the presence of unavoidable microscopic ﬂuctuations or external

perturbations, like small vibrations of the container. We have also seen that ir-

reversible processes are driving the system towards a unique equilibrium state

where the thermodynamic potentials take extremum values. In the particular

case of isolated systems, the unique equilibrium state is characterized by a

maximum value of the entropy. The fact of reaching or remaining in a state of

maximum or minimum potential makes that any equilibrium state be stable.

When internal ﬂuctuations or external perturbations drive the system away

from equilibrium, spontaneous irreversible processes will arise that bring the

system back to equilibrium. In the following sections, we will exploit the con-

sequences of equilibrium stability successively in single and multi-component

homogeneous systems.

1.6.1 Stability of Single Component Systems

Imagine a one-component system of entropy S, energy U, and volume V in

equilibrium and enclosed in an isolated container. Suppose that a hypotheti-

cal impermeable internal wall splits the system into two subsystems I and II

such that S = S

I

+S

II

, U = U

I

+U

II

= constant, V = V

I

+V

II

= constant. Un-

der the action of a disturbance, either internal or external, the wall is slightly

displaced and in the new state of equilibrium, the energy and volume of the

two subsystems will take the values U

I

+∆U, V

I

+∆V, U

II

−∆U, V

II

−∆V ,

respectively; let ∆S be the corresponding change of entropy. But at equilib-

rium, S is a maximum so that perturbations can only decrease the entropy

1.6 Stability of Equilibrium States 25

∆S < 0, (1.57)

while, concomitantly, spontaneous irreversible processes will bring the sys-

tem back to its initial equilibrium conﬁguration. Should ∆S > 0, then the

ﬂuctuations would drive the system away from its original equilibrium state

with the consequence that the latter would be unstable. Let us now explore

the consequences of inequality (1.57) and perform a Taylor-series expansion

of S(U

I

, V

I

, U

II

, V

II

) around the equilibrium state. For small perturbations,

we may restrict the developments at the second order and write symbolically

∆S = S −S

eq

= dS + d

2

S + < 0. (1.58)

From the property that S is extremum in equilibrium, the ﬁrst-order terms

vanish (dS = 0) and we are left with the calculation of d

2

S: it is found (as

detailed in Box 1.5) that

d

2

S = −T

2

C

V

(dT

−1

)

2

−

1

V Tκ

T

(dV )

2

< 0, (1.59)

where C

V

is the heat capacity at constant volume and κ

T

the isothermal

compressibility.

Box 1.5 Calculation of d

2

S for a Single Component System

Since the total energy and volume are constant dU

I

= −dU

II

= dU,

dV

I

= −dV

II

= dV , one may write

d

2

S =

1

2

_

_

∂

2

S

I

∂U

2

I

+

∂

2

S

II

∂U

2

II

_

eq

(dU)

2

+ 2

_

∂

2

S

I

∂U

I

∂V

I

+

∂

2

S

II

∂U

II

∂V

II

_

eq

dUdV

+

_

∂

2

S

I

∂V

2

I

+

∂

2

S

II

∂V

2

II

_

eq

(dV )

2

_

≤ 0. (1.5.1)

Recalling that the same substance occupies both compartments, S

I

and

S

II

and their derivatives will present the same functional dependence with

respect to the state variables, in addition, these derivatives are identical in

subsystems I and II because they are calculated at equilibrium. If follows

that (1.5.1) may be written as

d

2

S = S

UU

(dU)

2

+ 2S

UV

dUdV +S

V V

(dV )

2

≤ 0, (1.5.2)

wherein

S

UU

=

_

∂

2

S

∂U

2

_

V

=

_

∂T

−1

∂U

_

V

, S

V V

=

_

∂

2

S

∂V

2

_

U

=

_

∂(pT

−1

)

∂V

_

U

,

S

UV

=

_

∂

2

S

∂U∂V

_

V

=

_

∂T

−1

∂V

_

U

=

_

∂(pT

−1

)

∂U

_

V

. (1.5.3)

26 1 Equilibrium Thermodynamics: A Review

To eliminate the cross-term in (1.5.2), we replace the diﬀerential dU by

dT

−1

, i.e.

dT

−1

= S

UU

dU +S

UV

dV, (1.5.4)

whence

d

2

S =

1

S

UU

(dT

−1

)

2

+

_

S

V V

−

S

2

UV

S

UU

_

(dV )

2

> 0. (1.5.5)

Furthermore, since

S

UU

=

_

∂T

−1

∂U

_

V

= −

1

T

2

_

∂T

∂U

_

V

= −

1

T

2

C

V

(1.5.6)

and

S

V V

−

S

2

UV

S

UU

=

_

∂(pT

−1

)

∂V

_

T

=

1

T

_

∂p

∂V

_

T

= −

1

V Tκ

T

, (1.5.7)

as can be easily proved (see Problem 1.9), (1.5.5) becomes

d

2

S = −T

2

C

V

(dT

−1

)

2

−

1

V Tκ

T

(dV )

2

≤ 0. (1.5.8)

The criterion (1.59) for d

2

S < 0 leads to the following conditions of stability

of equilibrium:

C

V

= (¯ dQ

rev

/dT)

V

> 0, κ

T

= −(1/V )(∂V/∂p)

T

> 0. (1.60)

The ﬁrst criterion is generally referred to as the condition of thermal stability;

it means merely that, removing reversibly heat, at constant volume, must

decrease the temperature. The second condition, referred to as mechanical

stability, implies that any isothermal increase of pressure results in a diminu-

tion of volume, otherwise, the system would explode because of instabil-

ity. Inequalities (1.60) represent mathematical formulations of Le Chatelier’s

principle, i.e. that any deviation from equilibrium will induce a spontaneous

process whose eﬀect is to restore the original situation. Suppose for exam-

ple that thermal ﬂuctuations produce suddenly an increase of temperature

locally in a ﬂuid. From the stability condition that C

V

is positive, and heat

will spontaneously ﬂow out from this region (¯ dQ<0) to lower its tempera-

ture (dT <0). If the stability conditions are not satisﬁed, the homogeneous

system will evolve towards a state consisting of two or more portions, called

phases, like liquid water and its vapour. Moreover, when systems are driven

far from equilibrium, the state is no longer characterized by an extremum

principle and irreversible processes do not always maintain the system stable

(see Chap. 6).

1.6 Stability of Equilibrium States 27

1.6.2 Stability Conditions for the Other

Thermodynamic Potentials

The formulation of the stability criterion in the energy representation is

straightforward. Since equilibrium is characterized by minimum energy, the

corresponding stability criterion will be expressed as d

2

U(S, V ) ≥ 0 or, more

explicitly,

U

SS

≥ 0, U

V V

≥ 0, U

SS

U

V V

−(U

SV

)

2

≥ 0 (1.61)

showing that the energy is jointly a convex function of U and V (and also of

N in open systems).

The results are also easily generalized to the Legendre transformations

of S and U. As an example, consider the Helmholtz’s free energy F. From

dF = −S dT −p dV , it is inferred that

F

TT

= −T

−1

C

V

≤ 0, F

V V

=

1

V κ

T

≥ 0 (1.62)

from which it follows that F is a concave function of temperature and a convex

function of the volume as reﬂected by the inequalities (1.62). By concave

(convex) function is meant a function that lies everywhere below (above)

its family of tangent lines, be aware that some authors use the opposite

deﬁnition for the terms concave and convex. Similar conclusions are drawn

for the enthalpy, which is a convex function of entropy and a concave function

of pressure:

H

SS

≥ 0, H

pp

≤ 0. (1.63a)

Finally, the Gibbs’ free energy G is jointly a concave function of temperature

and pressure

G

TT

≤ 0, G

pp

≤ 0, G

TT

G

pp

−(G

Tp

)

2

≥ 0. (1.63b)

1.6.3 Stability Criterion of Multi-Component Mixtures

Starting from the fundamental relation of a mixture of n constituents in

the entropy representation, S = S(U, V, m

1

, m

2

, . . . , m

n

), it is detailed in

Box 1.6 that the second-order variation d

2

S, which determines the stability,

is given by

d

2

S = dT

−1

dU + d(pT

−1

)dV −

n

k=1

d(¯ µ

k

T

−1

)dm

k

≤ 0. (1.64)

At constant temperature and pressure, inequality (1.64) reduces to

n

k,l

∂¯ µ

k

∂m

l

dm

k

dm

l

≥ 0, (1.65)

28 1 Equilibrium Thermodynamics: A Review

Box 1.6 Calculation of d

2

S for Multi-Component Systems

In an N-component mixture of total energy U, total volume V , and total

mass m =

k

m

k

the second variation of S is

d

2

S =

1

2

(S

UU

)

eq

(dU)

2

+

1

2

(S

V V

)

eq

(dV )

2

+

1

2

k,l

(S

m

k

m

l

)

eq

dm

k

dm

l

+ (S

UV

)

eq

dUdV +

k

(S

Um

k

)

eq

dUdm

k

+

k

(S

V,m

k

)

eq

dV dm

k

.

(1.6.1)

Making use of the general results ∂S/∂U = 1/T, ∂S/∂V = p/T, ∂S/∂m

k

=

−µ

k

/T, the above expression can be written as

d

2

S =

1

2

_

∂T

−1

∂U

dU +

∂T

−1

∂V

dV +

k

∂T

−1

∂m

k

dm

k

_

dU

+

1

2

_

∂(pT

−1

)

∂U

dU +

∂(pT

−1

)

∂V

dV +

k

∂(pT

−1

)

∂m

k

_

dV

−

1

2

k

_

∂(¯ µ

k

T

−1

)

∂U

dU +

∂(¯ µ

k

T

−1

)

∂V

dV +

l

∂(¯ µ

k

T

−1

)

∂m

l

dm

l

_

dm

k

(1.6.2)

from which follows the general stability condition

d

2

S =

1

2

_

dT

−1

dU + d(pT

−1

)dV −

n

k=1

d(¯ µ

k

T

−1

)dm

k

_

≤ 0. (1.6.3)

to be satisﬁed whatever the values of dm

k

and dm

l

from which follows that:

∂¯ µ

k

∂m

k

≥ 0, det

¸

¸

¸

¸

∂¯ µ

k

∂m

l

¸

¸

¸

¸

≥ 0. (1.66)

The criteria (1.65) or (1.66) are referred to as the conditions of stability

with respect to diﬀusion. The ﬁrst inequality (1.66) indicates that the stabil-

ity of equilibrium implies that any increase on mass of a given constituent

will increase its chemical potential. This provides another example of the

application of Le Chatelier’s principle. Indeed, any non-homogeneity which

manifests in a part of the system in the form of increase of mass will induce

locally an increase of chemical potential. Since the latter is larger than its

ambient value, there will be a net ﬂow of matter from high to lower chemical

potentials that will tend to eradicate the non-homogeneity.

1.7 Equilibrium Chemical Thermodynamics 29

1.7 Equilibrium Chemical Thermodynamics

In the last part of this chapter, we shall apply the general results of equi-

librium thermodynamics to chemical thermodynamics. As an illustration,

consider the reaction of synthesis of hydrogen chloride

H

2

+ Cl

2

2HCl (1.67)

or more generally

n

k=1

ν

k

X

k

= 0, (1.68)

where the X

k

s are the symbols for the n chemical species and ν

k

the stoi-

chiometric coeﬃcients; conventionally the latter will be counted positive when

they correspond to products and negative for reactants. In the above example,

X

1

= H

2

, X

2

= Cl

2

, X

3

= HCl and ν

1

= −1, ν

2

= −1, ν

3

= 2, n = 3. The

reaction may proceed in either direction depending on temperature, pressure,

and composition; in equilibrium, the quantity of reactants that disappear is

equal to the quantity of products that instantly appear. The change in the

mole numbers dN

k

of the various components of (1.68) is governed by

dN

H

2

−1

=

dN

Cl

2

−1

=

dN

HCl

2

≡ dξ, (1.69)

where ξ is called the degree of advancement or extent of reaction. At the

beginning of the reaction ξ = 0; its time derivative dξ/dt is related to the

velocity of reaction which vanishes at chemical equilibrium (see Chap. 4).

The advantage of the introduction of ξ is that all the changes in the mole

numbers are expressed by one single parameter, indeed from (1.69),

dN

k

= ν

k

dξ, (1.70)

and, after integration,

N

k

= N

0

k

+ν

k

ξ, (1.71)

wherein superscript 0 denotes the initial state; observe that the knowledge

of ξ completely speciﬁes the composition of the system. When expressed in

terms of the mass of the constituents, (1.70) reads as dm

k

= ν

k

M

k

dξ with M

k

the molar mass of k; after summation on k, one obtains the mass conservation

law

n

k=1

ν

k

M

k

= 0. (1.72)

The above results are directly generalized when r chemical reactions are

taking place among the n constituents. In this case, (1.70) and (1.72) will,

respectively, be of the form

30 1 Equilibrium Thermodynamics: A Review

dN

k

=

r

j=1

ν

jk

dξ

j

(k = 1, 2, . . . , n),

n

k=1

ν

jk

M

k

= 0 (j = 1, 2, . . . , r).

(1.73)

In Sects. 1.7.1–1.7.3, we discuss further the conditions for chemical equilib-

rium and the consequences of stability of equilibrium.

1.7.1 General Equilibrium Conditions

Since chemical reactions take generally place at constant temperature and

pressure, it is convenient to analyse them in terms of Gibbs’ free energy

G = G(T, p, N

1

, N

2

, . . . , N

n

). At constant temperature and pressure, the

change in G associated with the variations dN

k

in the mole numbers is

dG =

n

k=1

µ

k

dN

k

, (1.74)

where µ

k

is the chemical potential per mole of species k. This µ

k

is closely

related to that appearing in (1.23), because the mass of the species k is

m

k

= M

k

N

k

, with M

k

the corresponding molar mass. Then, it is immediate

to see that µ

k

= ¯ µ

k

/M

k

. Since µ

k

and ¯ µ

k

are often found in the literature,

it is useful to be acquainted with both of them. After substitution of dN

k

by

its value (1.70), one has

dG =

_

n

k=1

ν

k

µ

k

_

dξ = −/dξ, (1.75)

wherein, with De Donder, we have introduced the “aﬃnity” of the reaction

as deﬁned by

/ = −

n

k=1

ν

k

µ

k

. (1.76)

Since G is a minimum at equilibrium (dG/dξ = 0), the condition of chemical

equilibrium is that the aﬃnity is zero:

/

eq

= 0. (1.77)

In presence of r reactions, equilibrium implies that the aﬃnity of each in-

dividual reaction vanishes: (/

j

)

eq

= 0(j = 1, 2, . . . , r). To better apprehend

the physical meaning of the result (1.77), let us express / in terms of phys-

ical quantities. In the case of ideal gases or diluted solutions, the chemical

potential of constituent k can be written as

µ

k

= µ

0

k

(p, T) +RT ln x

k

, (1.78)

1.7 Equilibrium Chemical Thermodynamics 31

where x

k

= N

k

/N is the mole fraction of substance k, N the total number

of moles and µ

0

k

(p, T) is the part of chemical potential depending only on p

and T. For non-ideal systems, the above form is preserved at the condition to

replace ln x

k

by ln(x

k

γ

k

) where γ

k

is called the activity coeﬃcient and a

k

=

γ

k

x

k

the activity. By substituting (1.78) in expression (1.76) of the aﬃnity,

we can express the equilibrium condition in terms of the mole fractions, which

are measurable quantities, and one has

/

eq

≡

k

(−ν

k

µ

0

k

) −RT

k

ν

k

ln x

k

= 0. (1.79)

Deﬁning the equilibrium constant K(T, p) by means of ln K(T, p) =

−(

k

ν

k

µ

0

k

)/RT, the previous relation can be cast in the simple form

/

eq

= RT ln

K(T, p)

x

ν

1

1

x

ν

2

2

x

ν

n

n

= 0, (1.80)

whence

x

ν

1

1

(ξ

eq

)x

ν

2

2

(ξ

eq

) x

ν

n

n

(ξ

eq

) = K(T, p), (1.81)

which is called the mass action law or Guldberg and Waage law. This is the

key relation in equilibrium chemistry: it is one algebraic equation involving

a single unknown, namely, the value of ξ

eq

of degree of advancement which

gives the corresponding number of moles in equilibrium through (1.71). If

K(T, p) is known as a function of T and p for a particular reaction, all the

equilibrium mole fractions can be computed by the mass action law. Going

back to our illustrative example (1.67) and assuming that each component is

well described by the ideal gas model, the law of mass action is

x

2

HCl

x

H

2

x

Cl

2

= K(T, p). (1.82)

When a number r of reactions are implied, we will have r algebraic equations

of the form (1.81) to be solved for the unknowns (ξ

1

)

eq

, (ξ

2

)

eq

, . . . , (ξ

r

)

eq

.

1.7.2 Heat of Reaction and van’t Hoﬀ Relation

Most of the chemical reactions supply or absorb heat, thus the heat of reaction

is an important notion in chemistry. To introduce it, let us start from the

ﬁrst law written as

¯ dQ = dH −V dp, (1.83)

with the enthalpy H given by the equation of state H = H(T, p, ξ), whose

diﬀerential form is

dH =

_

∂H

∂T

_

p,ξ

dT +

_

∂H

∂p

_

T,ξ

dp +

_

∂H

∂ξ

_

T,p

dξ. (1.84)

32 1 Equilibrium Thermodynamics: A Review

Substitution of (1.84) in (1.83) yields

¯ dQ = C

p,ξ

dT +h

T,ξ

dp −r

T,p

dξ, (1.85)

where C

p

= (∂H/∂T)

p,ξ

, h

T

= (∂H/∂p)

T,ξ

− V , and r

T,p

= −(∂H/∂ξ)

T,p

designate the speciﬁc heat at constant pressure, the heat compressibility and

the heat of reaction at constant temperature and pressure, respectively. The

heat of reaction is positive if the reaction is exothermic (which corresponds to

delivered heat) and negative if the reaction is endothermic (which corresponds

to absorbed heat). In terms of variation of aﬃnity with temperature, the heat

of reaction is given by

r

T,p

= −T

_

∂/

∂T

_

p,ξ

eq

. (1.86)

This is directly established by using the result

H = G+TS = G−T

_

∂G

∂T

_

p,ξ

. (1.87)

From the deﬁnition of the heat of reaction, one has

r

T,p

= −

_

∂G

∂ξ

_

T,p

+T

_

∂

∂T

_

∂G

∂ξ

_

T,p

_

p,ξ

, (1.88)

with (∂G/∂ξ)

T,p

= −/ in virtue of Gibbs’ equation (1.75), substituting this

result in (1.88) and recognizing that /

eq

= 0, one obtains (1.86). By means

of (1.78) of µ

k

and (1.79) of /, it is easily found from (1.86) that

r

T,p

= −RT

2

∂

∂T

ln K(T, p). (1.89)

This is the van’t Hoﬀ relation, which is very important in chemical ther-

modynamics; it permits to determine the heat of reaction solely from the

measurements of the equilibrium constant K(T, p) at diﬀerent temperatures.

1.7.3 Stability of Chemical Equilibrium

and Le Chatelier’s Principle

For the clarity of the presentation, let us recall the condition (1.65) for sta-

bility of diﬀusion which can be cast in the form:

n

k,l

µ

kl

dN

k

dN

l

> 0, µ

kl

=

∂µ

k

∂N

l

. (1.90)

1.7 Equilibrium Chemical Thermodynamics 33

Assuming that there are no simultaneous reactions and inserting (1.70), one

obtains

n

k,l

µ

kl

ν

k

ν

l

(dξ)

2

> 0. (1.91)

From the other side, it follows from the deﬁnition of the aﬃnity that

∂/

∂ξ

= −

n

k,l

ν

k

∂N

k

∂N

l

∂N

l

∂ξ

= −

n

k,l

µ

kl

ν

k

ν

l

, (1.92)

so that the criterion of stability (1.90) can be cast in the simple form

−

∂/

∂ξ

(dξ)

2

> 0 or

∂/

∂ξ

< 0. (1.93)

The above result is easily generalized for r simultaneous reactions:

r

m,n

∂/

m

∂ξ

n

dξ

m

dξ

n

< 0 (m, n = 1, 2, . . . , r). (1.94)

Perhaps one of the most interesting consequences of the stability criterion

(1.93) is in the form of Le Chatelier’s moderation principle. Let us ﬁrst ex-

amine the eﬀect on chemical equilibrium of a temperature change at constant

pressure. The shift of equilibrium with temperature is measured by the quan-

tity (∂ξ/∂T)

p,A

, index / is introduced here because in chemical equilibrium,

/ is constant, in fact zero. It is a mathematical exercise to prove that (see

Problem 1.8)

_

∂ξ

∂T

_

p,A

= −

(∂//∂T)

p,ξ

(∂//∂ξ)

p,T

. (1.95)

The numerator is related to the heat of reaction by (1.86) from which follows

that:

_

∂ξ

∂T

_

p,A

=

1

T

r

T,p

(∂//∂ξ)

p,T

. (1.96)

According to the stability condition (1.93), the denominator is a negative

quantity so that the sign of (∂ξ/∂T)

p,A

is opposite to the sign of the heat

of reaction r

T,p

. Therefore, an increase of temperature at constant pressure

will shift the reaction in the direction corresponding to endothermic reaction

(r

T,p

< 0). This is in the direction in which heat is absorbed, thus opposing

the increase of temperature. Similar results are obtained by varying the pres-

sure: an increase of pressure at constant temperature will cause the reaction

to progress in the direction leading to a diminution of volume, thus weak-

ening the action of the external eﬀect. These are particular examples of the

more general principle of Le Chatelier stating that any system in chemical

equilibrium undergoes, under the eﬀect of external stimuli, a compensating

change which will be always in the opposite direction.

34 1 Equilibrium Thermodynamics: A Review

1.8 Final Comments

Equilibrium thermodynamics constitutes a unique and universal formalism

whose foundations are well established and corroborated by experience. It

has also been the subject of numerous applications. It should nevertheless be

kept in mind that equilibrium thermodynamics is of limited range as it deals

essentially with equilibrium situations and idealized reversible processes. It is

therefore legitimate to ask to what extent equilibrium thermodynamics can

be generalized to cover more general situations as non-homogeneous systems,

far from equilibrium states and irreversible processes. Many eﬀorts have been

spent to meet such objectives and have resulted in the developments of various

approaches coined under the generic name of non-equilibrium thermodynam-

ics. It is our purpose in the forthcoming chapters to present, to discuss, and

to compare the most recent and relevant – at least in our opinion – of these

beyond of equilibrium theories.

There exists a multiplicity of excellent textbooks on equilibrium thermo-

dynamics and it would be unrealistic to go through the complete list. Let

us nevertheless mention the books by Callen (1985), Duhem (1911), Gibbs

(1948), Kestin (1968), Prigogine (1947), Kondepudi and Prigogine (1998) and

Zemansky (1968), which have been a source of inspiration for the present

chapter.

1.9 Problems

1.1. Mechanical work. Starting from the mechanical deﬁnition of work

¯ dW = F dx (scalar product of force and displacement), show that the

work done during the compression of a gas of volume V is ¯ dW = −p dV , and

that the same expression is valid for an expansion.

1.2. Carnot cycle. Show that the work performed by an engine during an

irreversible cycle operating between two thermal reservoirs at temperatures

T

1

and T

2

< T

1

is given by W = W

max

−T

2

∆S, where ∆S is the increase of

entropy of the Universe, and W

max

is the corresponding work performed in

a reversible Carnot cycle.

1.3. Fundamental relation. In the entropy representation, the fundamental

equation for a monatomic ideal gas is

S(U, V, N) =

N

N

0

S

0

+NRln

_

_

U

U

0

_

3/2

_

V

V

0

__

N

N

0

_

−5/2

_

,

with R the ideal gas constant, and the subscript 0 standing for an arbitrary

reference state. By using the formalism of equilibrium thermodynamics, show

that the thermal and caloric equations of state for this system are pV = NRT

and U =

3

2

NRT, respectively.

1.9 Problems 35

1.4. Maxwell’s relations. The equality of the second crossed derivatives of the

thermodynamic potentials is a useful tool to relate thermodynamic quantities.

Here, we will consider some consequences related to second derivatives of the

entropy. (a) Show that dS(T, V ) at constant N may be expressed as

dS =

1

T

_

∂U

∂T

_

V

dT +

1

T

__

∂U

∂V

_

T

+p

_

dV.

(b) By equating the second crossed derivatives of S in this expression, show

that

p +T

_

∂p

∂T

_

V

=

_

∂U

∂V

_

T

.

(c) Using this equation, show that for ideal gases, for which pV = NRT,

the internal energy does not depend on V , i.e. (∂U/∂V )

T

= 0. (d) For elec-

tromagnetic radiation, p = (1/3)(U/V ). Using this result and the relation

obtained in (b), show that for that system, one has U = aV T

4

. Remark:

This expression is closely related to the Stefan–Boltzmann law for the heat

ﬂux radiated by a black body, namely q = σT

4

, with σ the Stefan–Boltzmann

constant, indeed it follows from the relation q = 1/4c(U/V ), with c the speed

of light, and a given by a = 4σ/c.

1.5. Maxwell’s relations. Prove that c

p

− c

v

= Tvα

2

/κ

T

by making use of

Maxwell’s relations.

1.6. Van der Waals gases. The thermal equation of state for real gases was

approximated by van der Waals in the well-known expression

_

p +a

_

N

V

_

2

_

(V −bN) = NRT,

where a and b are positive constants, ﬁxed for each particular gas, which are,

respectively, related to the attractive and repulsive intermolecular forces and

are null for ideal gases. (a) Using the expressions derived in Problem 1.4,

show that the caloric equation of state U = U(T, V, N) has the form

U(T, V, N) = U

id

(T, N) −a

N

2

V

,

with U

id

(T, N) is the internal energy for ideal gases. (b) Calculate the change

of temperature in an adiabatic expansion against the vacuum, i.e. the vari-

ation of T in terms of V at constant internal energy. (c) Find the curve

separating the mechanically stable region from the mechanically unstable re-

gion in the plane p − V , the mechanical stability condition being given by

(1.60). (d) The maximum of such a curve, deﬁned by the additional con-

dition (∂

2

p/∂V

2

)

T,N

= 0, is called the critical point. Show that the val-

ues of p, V , and T at the critical point are p

c

= (1/27)(a/b

2

), V

c

= 3bN,

T

c

= (8/27)(a/bR), respectively.

36 1 Equilibrium Thermodynamics: A Review

1.7. Kelvin’s statement of the second law. Verify that in a reversible trans-

formation at ﬁxed temperature and volume, the maximum reversible work

delivered by the system is equal to the decrease of the Helmholtz’s free energy

−dF. As an aside result, show that it is not possible to obtain work from

an engine operating in a cycle and in contact with one single source of heat.

This result is known as Kelvin’s statement of the second law.

1.8. Mathematical relation. Verify the general result

_

∂y

∂x

_

f,z

= −

(∂f/∂x)

y,z

(∂f/∂y)

x,z

.

Hint: Consider f = f(X, Y, Z) = constant as an implicit function of the

variables X, Y, Z and write explicitly df = 0.

1.9. Stability coeﬃcient. Prove that (see Box 1.5)

∂

2

S

∂V

2

−

(∂

2

S/∂U∂V )

2

∂

2

S/∂U

2

=

1

T

_

∂p

∂V

_

T

.

Hint: Construct the Massieu function (i.e. Legendre transformation of en-

tropy), namely S[T

−1

] = S − T

−1

U. From the diﬀerential of S[T

−1

], derive

(∂

2

S[T

−1

]/∂V

2

)

T

.

1.10. Stability conditions. Reformulate the stability analysis of Sect. 1.6.1 by

considering that the total entropy and volume are kept constant and by

expanding the total energy in Taylor’s series around equilibrium.

1.11. Le Chatelier’s principle. Show that an increase in pressure, at constant

temperature, causes the chemical reaction to proceed in that direction which

decreases the total volume.

1.12. Le Chatelier’s principle. The reaction of dissociation of hydrogen iodide

2HI →H

2

+I

2

is endothermic. Determine in which direction equilibrium will

be shifted when (a) the temperature is decreased at constant pressure and

(b) the pressure is decreased at constant temperature.

Chapter 2

Classical Irreversible Thermodynamics

Local Equilibrium Theory of Thermodynamics

Equilibrium thermodynamics is concerned with ideal processes taking place

at inﬁnitely slow rate, considered as a sequence of equilibrium states. For

arbitrary processes, it may only compare the initial and ﬁnal equilibrium

states but the processes themselves cannot be described. To handle more

realistic situations involving ﬁnite velocities and inhomogeneous eﬀects, an

extension of equilibrium thermodynamics is needed.

A ﬁrst insight is provided by the so-called “classical theory of irreversible

processes” also named “classical irreversible thermodynamics” (CIT). This

borrows most of the concepts and tools from equilibrium thermodynamics

but transposed at a local scale because non-equilibrium states are usually in-

homogeneous. The objective is to cope with non-equilibrium situations where

basic physical quantities like mass, temperature, pressure, etc. are not only

allowed to change from place to place, but also in the course of time. As shown

in the present and the forthcoming chapters, this theory has been very use-

ful in dealing with a wide variety of practical problems. Pioneering works

in this theory were accomplished by Onsager (1931) and Prigogine (1961);

these authors were awarded the Nobel Prize in Chemistry in 1968 and 1977,

respectively. Other important and inﬂuential contributions are also found in

the works of Meixner and Reik (1959), de Groot and Mazur (1962), Gyarmati

(1970), and many others, which have enlarged the theory to a wider number

of applications and have clariﬁed its foundations and its limits of validity.

The principal aims of classical irreversible thermodynamics are threefold.

First to provide a thermodynamic support to the classical transport equations

of heat, mass, momentum, and electrical charge, as the Fourier’s law (1810)

relating the heat ﬂux to the temperature gradient, Fick’s relation (1850)

between the ﬂux of matter and the mass concentration gradient, Ohm’s equa-

tion (1855) between electrical current and potential, and Newton–Stokes’

law (1687, 1851) relating viscous pressure to velocity gradient in ﬂuids. A

second objective is to propose a systematic description of the coupling be-

tween thermal, mechanical, chemical, and electromagnetic eﬀects, as the Soret

(1879) and Dufour (1872) eﬀects, coupling heat, and mass transport, and the

37

38 2 Classical Irreversible Thermodynamics

Seebeck (1821), Peltier (1836) and Thomson eﬀects, coupling thermal trans-

port, and electric current. A third objective is the study of stationary non-

equilibrium dissipative states, whose properties do not depend on time, but

which are characterized by a non-homogeneous distribution of the variables

and non-vanishing values of the ﬂuxes.

The present chapter is divided in two parts: in the ﬁrst one, we recall

brieﬂy the general statements underlying the classical theory of irreversible

processes. The second part is devoted to the presentation of a few simple

illustrations, as heat conduction in rigid bodies, matter transport, and hydro-

dynamics. Chemical reactions and coupled transport phenomena, like ther-

moelectricity, thermodiﬀusion, and diﬀusion through a membrane are dealt

with separately in Chaps. 3 and 4.

2.1 Basic Concepts

The relevance of transport equations, which play a central role in non-

equilibrium thermodynamics – comparable, in some way, to equations of state

in equilibrium thermodynamics – justiﬁes some preliminary considerations.

Transport equations describe the amount of heat, mass, electrical charge, or

other quantities which are transferred per unit time between diﬀerent sys-

tems and diﬀerent regions of a system as a response to a non-homogeneity in

temperature T, molar concentration c, electric potential ϕ

e

. Historically, the

ﬁrst incursions into this subject are allotted to Fourier, Fick, and Ohm, who

proposed the nowadays well-known laws:

q = −λ∇T (Fourier’s law), (2.1)

J = −D∇c (Fick’s law), (2.2)

I = σ

e

∇ϕ

e

(Ohm’s law). (2.3)

Here q is the heat ﬂux (amount of internal energy per unit time and unit

area transported by conduction), J is the diﬀusion ﬂux (amount of matter,

expressed in moles, transported per unit time and unit area), and I is the

ﬂux of electric current (electric charge transported per unit time and area).

The coeﬃcients λ, D, and σ

e

are the thermal conductivity, diﬀusion coeﬃ-

cient, and electric conductivity, respectively. The knowledge of these various

transport coeﬃcients in terms of temperature, pressure, and mass concentra-

tion has important consequences in material sciences and more generally on

our everyday life. For instance, a low value of thermal conductivity is needed

for a better isolation of buildings; in contrast, large values are preferred to

avoid excessive heating of computers; the diﬀusion coeﬃcient is a fundamen-

tal parameter in biology and in pollution dispersal problems, while electrical

conductivity has a deep inﬂuence on the development and management of

electrical plants, networks, and microelectronic devices. The value of some

2.2 Local Equilibrium Hypothesis 39

transport coeﬃcients, as for instance electrical conductivity, may present

huge discontinuities as, for example, in superconductors, where conductiv-

ity diverges, thus implying a vanishing electrical resistivity, or in insulators,

where conductivity vanishes.

The physical content of the above transport laws is rather intuitive: heat

will ﬂow from regions with higher temperature to regions at lower temper-

ature and the larger is the temperature gradient the larger is the heat ﬂow.

Analogously, matter diﬀuses from regions with higher mass concentration to

regions with lower concentration, and electric positive charges move from

regions with higher electrical potential to regions with lower potential.

The evolution of a system in the course of time and space requires the

knowledge of the balance between the ingoing and the outgoing ﬂuxes. If the

outgoing ﬂow is larger than the incoming one, the amount of internal energy,

number of moles, or electric charge in the system will increase, and it will

decrease if the situation is reversed. When ingoing and outgoing ﬂuxes are

equal, the properties of the system will not change in the course of time, and

the system is in a non-equilibrium steady state. In equilibrium, all ﬂuxes van-

ish. It should be noticed that the above considerations apply only for so-called

conserved quantities, which means absence of production or consumption in-

side the system. When source terms are present, as for instance in chemical

reactions, one should add new contributions expressing the amount of moles,

which is produced or destroyed.

Expressions (2.1)–(2.3) of the classical transport laws were originally pro-

posed either from theoretical considerations or on experimental grounds. As

stated before, non-equilibrium thermodynamics aims to propose a general

scheme for the derivation of the transport laws by ensuring that they are

compatible with the laws of thermodynamics (for instance, thermal conduc-

tivity must be positive because, otherwise, heat would spontaneously ﬂow

from lower to higher temperature, in conﬂict with the second law). Indeed,

when λ, D, and σ

e

are scalar quantities it is relatively evident that they

must be positive; however, when they are tensors (as in anisotropic systems

or in the presence of magnetic ﬁelds) the thermodynamic restrictions on the

values of their components are not obvious and to obtain them a careful

and systematic analysis is required. It opens also the way to the study of

coupled situations where there are simultaneously non-homogeneities in tem-

perature, concentration, and electric potential, instead of considering these

eﬀects separately as in (2.1)–(2.3).

2.2 Local Equilibrium Hypothesis

The most important hypothesis underlying CIT is undoubtedly the local

equilibrium hypothesis. According to it, the local and instantaneous relations

between thermodynamic quantities in a system out of equilibrium are the same

40 2 Classical Irreversible Thermodynamics

as for a uniform system in equilibrium. To be more explicit, consider a system

split mentally in a series of cells, which are suﬃciently large for microscopic

ﬂuctuations to be negligible but suﬃciently small so that equilibrium is real-

ized to a good approximation in each individual cell. The size of such cells has

been a subject of debate, on which a good analysis can be found in Kreuzer

(1981) and Hafskjold and Kjelstrup (1995).

The local equilibrium hypothesis states that at a given instant of time,

equilibrium is achieved in each individual cell or, using the vocabulary of

continuum physics, at each material point. It should, however, be realized

that the state of equilibrium is diﬀerent from one cell to the other so that,

for example, exchanges of mass and energy are allowed between neighbouring

cells. Moreover, in each individual cell the equilibrium state is not frozen

but changes in the course of time. A better description of this situation is

achieved in terms of two timescales: the ﬁrst, τ

m

, denotes the equilibration

time inside one cell and it is of the order of the time interval between two

successive collisions between particles, i.e. 10

−12

s, at normal pressure and

temperature. The second characteristic time τ

M

is a macroscopic one whose

order of magnitude is related to the duration of an experiment, say about 1 s.

The ratio between both reference times is called the Deborah number De =

τ

m

/τ

M

. For De ¸1, the local equilibrium hypothesis is fully justiﬁed because

the relevant variables evolve on a large timescale τ

M

and do practically not

change over the time τ

m

, but the hypothesis is not appropriate to describe

situations characterized by De ¸1. Large values of De are typical of systems

with long relaxation times, like polymers, for which τ

m

may be of the order

of 100 s, or of high-frequency or very fast phenomena, such as ultrasound

propagation, shock waves, nuclear collisions, for which τ

M

is very short, say

between 10

−5

and 10

−10

s.

A ﬁrst consequence of the local equilibrium assumption is that all the

variables deﬁned in equilibrium as entropy, temperature, chemical potential,

etc. are univocally deﬁned outside equilibrium, but they are allowed to vary

with time and space. Another consequence is that the local state variables

are related by the same state equations as in equilibrium. This means, in

particular, that the Gibbs’ relation between entropy and the state variables

remains locally valid for each value of the time t and the position vector r.

For example, in the case of a n-component ﬂuid of total mass m, the local

Gibbs’ equation will be written as

ds = T

−1

du +pT

−1

dv −T

−1

n

k=1

µ

k

dc

k

, (2.4)

where s is the speciﬁc entropy (per unit mass), u is the speciﬁc internal

energy, related to the speciﬁc total energy e by u = e −

1

2

v v, with v the

velocity ﬁeld of the centre of mass of the cell, T is the absolute temperature,

p is the hydrostatic pressure, v is the speciﬁc volume related to the mass

density ρ by v = 1/ρ, c

k

= m

k

/m, the mass fraction of substance k, and µ

k

its chemical potential.

2.3 Entropy Balance 41

A third consequence follows from the property that, locally, the system

is stable. Therefore, in analogy with equilibrium situations, such quantities

as heat capacity, isothermal compressibility or the Lam´e coeﬃcients in the

theory of elasticity are positive deﬁnite.

More generally, Gibbs’ equation will take the form

ds(r, t) =

i

Γ

i

(r, t) da

i

(r, t), (2.5)

where a

i

(r, t) is an extensive state variable, like u, v, c

k

, while Γ

i

(r, t) is the

corresponding conjugate intensive state variable, for instance T, p or µ

k

.

The above relation is assumed to remain valid when expressed in terms of

the material (or Lagrangian) time derivative d/dt = ∂/∂t + v ∇, i.e. by

following a small cell moving with velocity v:

ds(r, t)

dt

=

i

Γ

i

(r, t)

da

i

(r, t)

dt

. (2.6)

From the kinetic theory point of view, the local equilibrium hypothesis is jus-

tiﬁed only for conditions where the Maxwellian distribution is approximately

maintained. Otherwise, it should be generalized, as indicated in the second

part of this book.

2.3 Entropy Balance

An important question is whether a precise deﬁnition can be attached to

the notion of entropy when the system is driven far from equilibrium. In

equilibrium thermodynamics, entropy is a well-deﬁned function of state only

in equilibrium states or during reversible processes. However, thanks to the

local equilibrium hypothesis, entropy remains a valuable state function even

in non-equilibrium situations. The problem of the deﬁnition of entropy and

corollary of intensive variables as temperature will be raised as soon as the

local equilibrium hypothesis is given up.

By material body (or system) is meant a continuum medium of total mass

m and volume V bounded by a surface Σ. Consider an arbitrary body, outside

equilibrium, whose total entropy at time t is S. The rate of variation of this

extensive quantity may be written as the sum of the rate of exchange with

the exterior d

e

S/dt and the rate of internal production, d

i

S/dt:

dS

dt

=

d

e

S

dt

+

d

i

S

dt

. (2.7)

As in Chap. 1, we adopt the convention that any quantity (like mass,

energy, entropy) is counted positive if supplied to the system and negative if

transferred from it to the surroundings. The quantity Td

i

S/dt is sometimes

42 2 Classical Irreversible Thermodynamics

called the uncompensated heat or the rate of dissipation. For further purpose,

it is convenient to introduce the notion of entropy ﬂux J

s

, i.e. the entropy

crossing the boundary surface per unit area and unit time, and the rate of

entropy production σ

s

, i.e. the entropy produced per unit volume and unit

time inside the system. In terms of these quantities, d

e

S/dt and d

i

S/dt may

be written as

d

e

S

dt

= −

_

Σ

J

s

n dΣ, (2.8)

d

i

S

dt

=

_

V

σ

s

dV , (2.9)

in which n is the unit normal pointing outwards the volume of the body.

Once entropy is deﬁned, it is necessary to formulate the second law, i.e.

to specify which kinds of behaviours are admissible in terms of the entropy

behaviour. The classical formulation of the second law due to Clausius states

that, in isolated systems, the possible processes are those in which the entropy

of the ﬁnal equilibrium state is higher or equal (but not lower) than the

entropy of the initial equilibrium state. In the classical theory of irreversible

processes, one introduces an even stronger restriction by requiring that the

entropy of an isolated system must increase everywhere and at any time, i.e.

dS/dt ≥ 0. In non-isolated systems, the second law will take the more general

form

d

i

S/dt > 0 (for irreversible processes), (2.10a)

d

i

S/dt = 0 (for reversible processes or at equilibrium). (2.10b)

It is important to realize that inequality (2.10a) does nor prevent that open or

closed systems driven out of equilibrium may be characterized by dS/dt < 0;

this occurs for processes for which d

e

S/dt < 0 and larger in absolute value

than d

i

S/dt. Several examples are discussed in Chap. 6.

After introducing the speciﬁc entropy s in such a way that S =

_

ρs dV

and using the deﬁnitions (2.8) and (2.9), the entropy balance (2.7) reads as

d

dt

_

ρs dV = −

_

Σ

J

s

n dΣ +

_

V

σ

s

dV . (2.11)

In virtue of the Gauss and Reynolds theorems, the above equation takes the

form

_

ρ

ds

dt

dV = −

_

V

∇ J

s

dV +

_

V

σ

s

dV , (2.12)

where ∇ ≡ (∂/∂x, ∂/∂y, ∂/∂z) designates the nabla operator whose compo-

nents are the partial space derivatives in Cartesian coordinates. Assuming

that (2.12) is valid for any volume V and that the integrands are continuous

functions of position, one can write the following local balance relation

ρ

ds

dt

= −∇ J

s

+σ

s

, (2.13)

2.3 Entropy Balance 43

with, in virtue of the second law as stated by ((2.10a) and (2.10b)),

σ

s

≥ 0, (2.14)

where the equality sign refers to reversible processes. This quantity is impor-

tant in engineering because the product Tσ

s

is a measure of the degradation

or dissipation of energy in engines, and its minimization may be useful to

enhance their eﬃciency.

Finally, it is worth to mention that the second law introduces an asym-

metry in time for irreversible processes, which is often known as an arrow of

time, an interesting physical and even philosophical concept that is brieﬂy

presented in Box 2.1.

Box 2.1 Irreversibility and the Arrow of Time

An interesting aspect of the principle of entropy increase is the special light

it sheds on the concept of time, a view that departs radically from classical

mechanics. The equations of Newtonian mechanics are deterministic and

reversible with respect to time reversal. Time is considered as an external

parameter, which describes the chronology of a succession of events; it is the

time given by our watch. Reversible processes do not distinguish between

the future and the past. In contrast, the principle of increase of entropy

makes possible the distinction between future and past as it implies an arrow

of time, which imposes that irreversible processes proceed spontaneously

within a given direction in time. It establishes a fundamental anisotropy

in Nature and provides a criterion allowing to decide whether a process is

going forwards or backwards.

In relation with the notions of time, reversibility and irreversibility

was raised the following problem. At the microscopic level, the motion of

the individual particles composing the macroscopic systems is described

by Newton’s equation and is therefore of reversible nature, whereas at the

macroscopic level, the systems behave irreversibly. The antagonism between

these two behaviours has been a major source of debate since the micro-

scopic interpretation of entropy by Boltzmann, in the years 1870. A wide-

spread view is that the behaviour of the systems is intrinsically reversible

but that the large number of particles makes that actually, it evolves towards

an irreversible dynamics due to our inability to follow each individual parti-

cle. In other words, irreversibility is an illusion raised up by our ignorance.

Following this attitude, irreversibility should not be related to the system

itself but to the observer.

Irreversibility is also associated with loss of information, a view rein-

forced by the information theory. Accordingly, the entropy is interpreted as

a lack of information about the microscopic state of the system, and loss

of information means higher entropy. When entropy production is large,

much information is lost per unit time, whereas in reversible processes no

information is lost. The strong interrelations between information theory

44 2 Classical Irreversible Thermodynamics

and thermodynamics have been underlined in many books and papers (e.g.

Shannon 1948; Jaynes 1963; Keizer 1987).

In the 1970s, the study of deterministic chaos has opened new per-

spectives. Accordingly, though the equations of mechanics are reversible

and deterministic, the sensibility of their solutions with respect to slight

changes of the initial conditions or by varying some external parameters as

temperature gradients (see Chap. 6) makes that the predictability is lost.

Interesting conceptual discussions on this subject have been put forward

by Prigogine (e.g. Prigogine and Stengers 1979; Prigogine 1996), who ar-

gues that irreversibility is present at all levels of description, microscopic

and macroscopic, and merges out, as a source of order, from the instabil-

ities present in the system. The breaking of time symmetry is introduced

by constructing an appropriate operator playing the role of entropy and

monotonically increasing in time. The formalism is coined under the name

of subdynamics. Prigogine’s line of thought is not unanimously accepted

and target of keen discussions (see, e.g. Thom 1972; Bricmont 1995; Van

Kampen 2000).

The above considerations show that the problem of evolution of re-

versibility towards irreversibility has not received a deﬁnitive answer yet

(Lebowitz 1999). A last remark will concern ageing, which is a sensible

problem to everybody and assuredly the most visible irreversible face of

life. Experiences of our daily life tell us that youth will not return. There

are many theories of biological ageing but these topics are outside the scope

of the present book. It seems rather evident that ageing is related to the

entropy production, or the amount of irreversibility, inside the metabolic

processes, and that small entropy production should produce slow ageing.

More considerations about the convergence of non-equilibrium thermody-

namics and biology may be found in Chap. 4.

2.4 General Theory

Our main objective is to present a macroscopic description of irreversible

processes and our major task is to determine the evolution equations of the

relevant local variables, in accordance with the fundamental laws of thermo-

dynamics. To give a general idea of the contents and assumptions underlying

classical irreversible thermodynamics, we consider a system outside equilib-

rium and characterized by a ﬁnite set of variables to be speciﬁed later on.

To make clear the general structure, we will proceed in several steps. The

same steps, as introduced in this section, will be systematically repeated in

the future.

2.4 General Theory 45

Step 1. Space of state variables

The choice of state variables is determined by three arguments:

1. It depends on the system under consideration and the degree of accuracy

that one wishes to achieve.

2. The variables must be “dynamically admissible”, which means that their

timescale is suﬃciently diﬀerent from the timescales corresponding to the

more microscopic levels of description.

3. Theoretical predictions should be in agreement with experimental obser-

vations, and in addition, the results should be reproducible. This does not

mean that the state variables are directly measurable but it is required to

express them in terms of measured quantities.

According to the local equilibrium hypothesis, the space of the state vari-

ables is the ensemble a(r, t) = [a

1

(r, t), a

2

(r, t), . . . , a

N

(r, t)] of the exten-

sive thermodynamic variables appearing in Gibbs’ equation, plus the velocity

ﬁeld v(r, t), i.e. the union of a mechanical variable (the barycentric velocity

of each subcell) and thermodynamic variables (i.e. each subcell is considered

as a thermodynamic system described by the usual set of thermodynamic

quantities as in equilibrium). In the case of an n-component ﬂuid, the ther-

modynamic variables are simply the speciﬁc volume v(r, t), the mass fraction

c

k

(r, t)(k = 1, 2, . . ., n) of the n constituents and the internal energy u(r, t).

Step 2. Evolution equations

They take the form of balance equations written in whole generality as

ρ

da

dt

= −∇ J

a

+σ

a

, (2.15)

where J

a

is the ﬂux term expressing the exchange with the surroundings and

σ

a

is the corresponding source term; when this term is zero, a is conserved.

To be explicit, and to identify the ﬂux and source terms, let us write down the

balance equations of total mass, mass concentrations, energy and momentum

for a n-constituent mixture (de Groot and Mazur 1962):

ρ

dv

dt

= ∇ v, (2.16)

ρ

dc

k

dt

= −∇ J

k

+σ

k

, (2.17)

ρ

du

dt

= −∇ q −P

T

: ∇v +ρr, (2.18)

ρ

dv

dt

= −∇ P+ρF, (2.19)

where the superscript T means transposition. In (2.16), the ﬂux term is simply

the velocity and there is no source because of conservation of mass. In (2.17)–

(2.19), the ﬂuxes are the diﬀusion ﬂux J

k

of substance k, the heat ﬂux q

and the pressure tensor P, respectively; the corresponding sources are the

46 2 Classical Irreversible Thermodynamics

rate of production σ

k

of substance k by chemical reactions, the mechanical

work −P

T

: (∇v) plus some possible energy source r per unit mass and the

external body force F per unit mass. The main diﬀerence between q and

r is that q measures the transfer of heat across the bounding surface while

r is the energy supply distributed within the volume. We refer to P as the

pressure tensor because pure hydrostatic pressure corresponds to P = pI,

where I is the identity tensor with components δ

ij

(δ

ij

= 1 for i = j and

δ

ij

= 0 for i ,= j); the colon in (2.18) stands for the double scalar product

A : B = A

ij

B

ji

. Einstein’s summation convention on repeated indices, when

they refer to Cartesian coordinates, will be used throughout the book.

The set (2.16)–(2.19) describes the evolution of the system, i.e. how the

basic properties change with time at every point in space. Note that it con-

tains more unknown quantities than equations, because the ﬂuxes q and P

are generally not known and have to be expressed in terms of the basic vari-

ables to close the system. In that respect the second law is very useful as

shown by the next step.

Step 3. Entropy production and second law of thermodynamics

It is the purpose to examine the consequences resulting from the entropy

inequality σ

s

> 0. Substitution of the evolution equations (2.15) of the basic

variables in the Gibbs’ relation (2.6) leads to an explicit equation for the

entropy evolution which, after comparison with the general expression (2.13)

of the entropy balance, allows us to identify J

s

and σ

s

. The latter is found

to consist of a sum of products of so-called thermodynamic ﬂuxes J

α

and

thermodynamic forces X

α

:

σ

s

=

α

J

α

X

α

. (2.20)

Note that the thermodynamic forces are not forces in the mechanical sense,

but they are quantities generally related to the gradients of the intensive

variables whereas the ﬂuxes J

α

can be identiﬁed with the ﬂuxes of energy,

mass, momentum, etc. In (2.20), the ﬂuxes and forces may be scalars, vectors,

or tensors, and the product J

α

X

α

stands indiﬀerently for the usual product

between two scalars, the scalar product between two vectors and the double

scalar product between two tensors. Explicit expressions for J

α

and X

α

will be

given later on while treating applications. It is important to point out that at

equilibrium (or for reversible processes), the thermodynamic ﬂuxes and forces

vanish identically so that entropy production is zero in such situations, as it

should.

Step 4. Linear ﬂux–force relations

Experience indicates that the thermodynamic ﬂuxes and forces are not inde-

pendent but that there exists a relationship between them. Moreover, it has

been observed that for a large class of irreversible processes, the ﬂuxes are

linear functions of the forces, to a good approximation. This is furthermore

the simplest way to ensure that the rate of entropy production is a positive

quantity. Within the hypothesis of linearity, one has

2.4 General Theory 47

J

α

=

β

L

αβ

X

β

. (2.21)

These ﬂux–force relations are named phenomenological relations, constitu-

tive or transport equations, they express the relation between causes (the

forces) and eﬀects (the ﬂuxes) and the speciﬁc properties of the materials

under study. At equilibrium, both members of (2.21) vanish identically. The

phenomenological coeﬃcients L

αβ

are generally depending on the intensive

variables T, p, and c

k

; the coeﬃcient L

αα

connects a ﬂow J

α

to its conjugate

force X

α

while L

αβ

describes the coupling between two irreversible processes

α and β. In thermoelectricity for instance, L

αα

is related to the electrical

resistance and L

αβ

to the coupling between the electric current and the heat

ﬂow. Simple examples of linear phenomenological relations are the Fourier’s,

Fick’s, and Ohm’s equations (2.1)–(2.3).

Step 5. Restrictions due to material symmetry: Curie’s law

According to relation (2.21), one should in principle be allowed to couple any

ﬂux to any force. However, material symmetry reduces the number of cou-

plings between ﬂuxes and forces. This property is known in CIT as Curie’s

law. It reﬂects the property that macroscopic causes cannot have more ele-

ments of symmetry than the eﬀects they produce. This restriction plays an

important role, especially in isotropic systems, for which the properties at

equilibrium are the same in all directions.

Truly, as shown in Box 2.2, it is evident that Curie’s law is nothing

more than an application of the representation theorems of isotropic tensors

(Spencer and Rivlin 1959; Truesdell and Noll 1965). To illustrate this state-

ment, consider an isotropic body subject to a hypothetical irreversible process

described by the ﬂuxes j (a scalar), J (a vector), T (a symmetric tensor of

order 2). In addition, the ﬂuxes j and J are assumed to depend on the ther-

modynamic forces x (a vector) and X(a tensor of order 2) but Tdepends only

on X.

As a consequence of isotropy, and under the hypothesis of linear ﬂux–force

relations (see Box 2.2), one obtains the following phenomenological relations:

j = l tr X, (2.22a)

J = A

1

x, (2.22b)

T = B

1

(tr X)I +B

2

X, (2.22c)

where the phenomenological scalar coeﬃcients l, A

1

, B

1

, and B

2

are inde-

pendent of x and X, and tr X denotes the trace of tensor X. The results

((2.22a), (2.22b), and (2.22c)) exhibit the property that, in isotropic systems

and within the linear regime, it is forbidden to couple ﬂuxes and forces of

diﬀerent tensorial character. For instance a chemical aﬃnity (a scalar) can-

not give raise to a heat ﬂux (a vector), similarly a temperature gradient

(a vector) is unable to induce a mechanical stress (a tensor of order 2).

48 2 Classical Irreversible Thermodynamics

Box 2.2 Curie’s Law

The existence of spatial symmetries in a material system contributes to sim-

plify the scheme of the phenomenological relations. Because of invariance of

the phenomenological equations under special orthogonal transformations,

some couplings between ﬂuxes and forces are not authorized in isotropic

systems. In the examples treated in this chapter, no ﬂuxes and forces of ten-

sorial order higher than the second will occur. We shall therefore consider

an isotropic material characterized by the three following phenomenological

relations

j = j(x, X), J = J(x, X), T = T(X), (2.2.1)

which contain ﬂuxes and forces of tensorial order 0 (the scalar j), order

1 (the vectors J and x) and order 2 (the symmetric tensors T and X).

Isotropy imposes that the above relations transform as follows under an

orthogonal transformation Q:

j(Q x; Q X Q

T

) = j(x, X), (2.2.2)

J(Q x; Q X Q

T

) = Q J(x, X), (2.2.3)

T(Q X Q

T

) = Q T(X) Q

T

. (2.2.4)

According to the theorems of representation of isotropic tensors (e.g.

Spencer and Rivlin 1959), the functions j, J, and T are isotropic if and

only if

j(x, X) = j[I

X

, II

X

, III

X

, x x, x (X x), x (X X x)], (2.2.5)

J(x, X) = (A

1

I +A

2

X+A

3

X X) x, (2.2.6)

T(X) = B

1

I +B

2

X+B

3

X X, (2.2.7)

the ﬂux j and the coeﬃcients A

i

(i = 1, 2, 3) are isotropic scalar functions

of x and X and the B

i

are isotropic scalars of X alone; I

X

, II

X

, and III

X

are the principal invariants of the tensor X, namely

I

X

= tr X, II

X

=

1

2

[I

2

X

−tr(X X)], III

X

= det X. (2.2.8)

By restricting the analysis to linear laws, (2.2.5)–(2.2.7) will take the simple

form

j = l(tr X), J = A

1

x, T = B

1

(tr X)I +B

2

X, (2.2.9)

wherein l, A

1

, B

1

, and B

2

are now scalars independent of x and X. Re-

lations (2.2.9) exhibit the property that in linear constitutive equations,

spatial symmetry allows exclusively the coupling between ﬂuxes and forces

of the same tensorial order, as concluded from Curie’s law.

2.4 General Theory 49

Step 6. Restrictions on the sign of phenomenological coeﬃcients

A direct restriction on the sign of the phenomenological coeﬃcients arises as a

consequence of the second law. Substitution of the linear ﬂux–force relations

(2.21) into (2.20) of the rate of entropy production yields the quadratic form

σ

s

=

αβ

L

αβ

X

α

X

β

≥ 0. (2.23)

According to standard results in algebra, the necessary and suﬃcient condi-

tions for σ

s

≥ 0 are that the determinant [L

αβ

+L

βα

[ and all its principal

minors are non-negative. It follows that

L

αα

≥ 0, (2.24)

while the cross-coeﬃcients L

αβ

must satisfy

L

αα

L

ββ

≥

1

4

(L

αβ

+L

βα

)

2

. (2.25)

In virtue of inequality (2.24), all the transport coeﬃcients like the heat con-

ductivity, the diﬀusion coeﬃcient, and the electrical resistance are positive,

meaning that heat ﬂows from high to low temperature, electrical current

from high to low electric potential, and neutral solutes from higher to lower

concentrations.

Step 7. Restrictions on L

αβ

due to time reversal: Onsager–Casimir’s

reciprocal relations

It was established by Onsager (1931) that, besides the restrictions on the sign,

the phenomenological coeﬃcients verify symmetry properties. The latter were

presented by Onsager as a consequence of “microscopic reversibility”, which

is the invariance of the microscopic equations of motion with respect to time

reversal t →−t. Accordingly, by reversing the time, the particles retrace their

former paths or, otherwise stated, there is a symmetry property between

the past and the future. Invoking the principle of microscopic reversibility

and using the theory of ﬂuctuations, Onsager was able to demonstrate the

symmetry property

L

αβ

= L

βα

. (2.26)

In Chaps. 4 and 11, we will present detailed derivations of (2.26).

It should, however, be stressed that the above result holds true only for

ﬂuctuations a

α

(t) = A

α

(t) −A

eq

α

of extensive state variables A

α

with respect

to their equilibrium values, which are even functions of time a

α

(t) = a

α

(−t).

In the case of odd parity of one of the variables, α or β, for which a

α

(t) =

−a

α

(−t), the coeﬃcients L

αβ

are skew-symmetric instead of symmetric, as

shown by Casimir (1945), i.e.

L

αβ

= −L

βα

. (2.27)

In a reference frame rotating with angular velocity ω and in presence of an

external magnetic ﬁeld H, Onsager–Casimir’s reciprocal relations take the

form

50 2 Classical Irreversible Thermodynamics

L

αβ

(ω, H) = ±L

βα

(−ω, −H), (2.28)

as it will be shown on dynamical bases in Chap. 11. The validity of the

Onsager–Casimir’s reciprocal relations is not limited to phenomenological

transport coeﬃcients that are scalar quantities as discussed earlier. Consider

for example an irreversible process taking place in an anisotropic crystal, such

that

J

α

=

β

L

αβ

X

β

, (2.29)

where ﬂuxes and forces are vectors and L

αβ

is a tensor of order 2. The

Onsager–Casimir’s reciprocal relations write now as

L

αβ

= ±(L

βα

)

T

. (2.30)

Transformation properties of the reciprocal relations have been discussed by

Meixner (1943) and Coleman and Truesdell (1960).

At ﬁrst sight, Onsager–Casimir’s reciprocal relations may appear as a

rather modest result. Their main merit is to have evidenced symmetry proper-

ties in coupled irreversible processes. As illustration, consider heat conduction

in an anisotropic crystal. The reciprocity relations imply that a temperature

gradient of 1

◦

Cm

−1

along the x-direction will give raise to a heat ﬂux in

the normal y-direction, which is the same as the heat ﬂow generated along

the x-axis by a temperature gradient of 1

◦

Cm

−1

along y. Another advan-

tage of the Onsager–Casimir’s reciprocal relations is that the measurement

(or the calculation) of a coeﬃcient L

αβ

alleviates the repetition of the same

operation for the reciprocal coeﬃcient L

βα

; this is important in practice as

the cross-coeﬃcients are usually much smaller (of the order of 10

−3

to 10

−4

)

than the direct coupling coeﬃcients, and therefore diﬃcult to measure or

even to detect.

Although the proof of the Onsager–Casimir’s reciprocal relations was

achieved at the microscopic level of description and for small deviations of

ﬂuctuations from equilibrium, these symmetry properties have been widely

applied in the treatment of coupled irreversible processes taking place at the

macroscopic scale even very far from equilibrium. It should also be kept in

mind that the validity of the reciprocity properties is secured as far as the

ﬂux–force relations are linear, but that they are not of application in the

non-linear regime.

2.5 Stationary States

Stationary states play an important role in continuum physics; they are de-

ﬁned by the property that the state variables, including the velocity, remain

unchanged in the course of time. For instance, if heat is supplied at one end

of a system and removed at the other end at the same rate, the temperature

at each point will not vary in time but will change from one position to the

2.5 Stationary States 51

other. Such a state cannot be confused with an equilibrium state, which is

characterized by a uniform temperature ﬁeld, no heat ﬂow, and a zero entropy

production. It is to be emphasized that the evolution of a system towards an

equilibrium state or a steady state is conditioned by the nature of the bound-

ary conditions. Since in a stationary state, entropy does not change in the

course of time, we can write in virtue of (2.7) that

−

¸

¸

¸

¸

dS

dt

¸

¸

¸

¸

out

+

¸

¸

¸

¸

dS

dt

¸

¸

¸

¸

in

+

d

i

S

dt

= 0, (2.31)

since the rate of entropy production is positive, it is clear that the entropy

delivered by the system to the external environment is larger than the entropy

that is entering. Using the vocabulary of engineers, the system degrades the

energy that it receives and this degradation is the price paid to maintain a

stationary state.

Stationary states are also characterized by interesting extremum principles

as demonstrated by Prigogine (1961): the most important is the principle of

minimum entropy production, which is discussed further. The importance of

variational principles has been recognized since the formulation of Hamilton’s

least action principle in mechanics stating that the average kinetic energy

less the average potential energy is minimum along the path of a particle

moving from one point to another. Quoting Euler, “since the construction

of the universe is the most perfect possible, being the handy work of an all-

wise Maker, nothing can be met in the world in which some minimum or

maximum property is not displayed”. It is indeed very attractive to believe

that a whole class of processes is governed by a single law of minimum or

maximum. However, Euler’s enthusiasm has to be moderated, as most of the

physical phenomena cannot be interpreted in terms of minima or maxima. In

equilibrium thermodynamics, maximization of entropy for isolated systems or

minimization of Gibbs’ free energy for systems at constant temperature and

pressure, for instance, provide important examples of variational principles.

Out of equilibrium, such variational formulations are much more limited.

For this reason, we pay here a special attention to the minimum entropy

production theorem, which is the best known among the few examples of

variational principles in non-equilibrium thermodynamics.

2.5.1 Minimum Entropy Production Principle

Consider a non-equilibrium process, for instance heat conduction or thermod-

iﬀusion taking place in a volume V at rest subjected to time-independent

constraints at its surface. The state variables a

1

, a

2

, . . . , a

n

are assumed to

obey conservation laws of the form

ρ

∂a

α

∂t

= −∇ J

α

(α = 1, 2, . . . , n), (2.32)

52 2 Classical Irreversible Thermodynamics

where ∂/∂t is the partial or Eulerian time derivative; processes of this kind,

characterized by absence of global velocity, are called purely dissipative. We

have seen in Sect. 2.4 that the total entropy P produced inside the system

can be written as

P =

_

α,β

L

αβ

X

α

X

β

dV. (2.33)

Since the thermodynamic forces take usually the form of gradients of intensive

variables

X

α

= ∇Γ

α

, (2.34)

where Γ

α

designates an intensive scalar variable, (2.33) becomes

P =

_

α,β

L

αβ

∇Γ

α

∇Γ

β

dV. (2.35)

We now wish to show that the entropy production is minimum in the sta-

tionary state. Taking the time derivative of (2.35) and supposing that the

phenomenological coeﬃcients are constant and symmetric, one obtains

dP

dt

= 2

_

α,β

L

αβ

∇Γ

α

∇

_

∂Γ

β

∂t

_

dV. (2.36)

After integration by parts and recalling that the boundary conditions are

time independent, it is found that

dP

dt

= −2

_

β

∂Γ

β

∂t

∇

_

α

L

αβ

∇Γ

α

_

dV

= −2

_

β

∂Γ

β

∂t

∇ J

β

dV = 2

_

ρ

β

∂Γ

β

∂t

∂a

β

∂t

dV, (2.37)

wherein use has been made successively of the linear ﬂux–force relations

(2.21) and the conservation law (2.32). In the stationary state, for which

∂α

β

/∂t = 0, one has

dP

dt

= 0. (2.38)

During the transient regime, (2.37) can be written as

dP

dt

=

_

α,β

∂a

α

∂Γ

β

∂Γ

α

∂t

∂Γ

β

∂t

dV, (2.39)

and since the ∂α

α

/∂Γ

β

terms (which represent, for instance, minus the heat

capacity or minus the coeﬃcient of isothermal compressibility) are negative

quantities because of stability of equilibrium, one may conclude that

dP

dt

≤ 0. (2.40)

2.5 Stationary States 53

This result proves that the total entropy production P decreases in the course

of time and that it reaches its minimum value in the stationary state. An

important aside result is that stationary states with a minimum entropy pro-

duction are unconditionally stable. Indeed, after application of an arbitrary

disturbance in the stationary state, the system will move towards a transitory

regime with a greater entropy production. But as the latter can only decrease,

the system will go back to its stationary state, which is therefore referred to

as stable. It is also worth to mention that P, a positive deﬁnite functional

with a negative time derivative, provides an example of Lyapounov’s function

(Lyapounov 1966), whose occurrence is synonymous of stability, as discussed

in Chap. 6.

It should, however, be emphasized that the above conclusions are far

from being general, as their validity is subordinated to the observance of

the following requirements:

1. Time-independent boundary conditions

2. Linear phenomenological laws

3. Constant phenomenological coeﬃcients

4. Symmetry of the phenomenological coeﬃcients

In practical situations, it is frequent that at least one of the above restrictions

is not satisﬁed, so that the criterion of minimum entropy production is of weak

bearing. It follows also that most of the stationary states met in the nature

are not necessarily stable as conﬁrmed by our everyday experience.

The result (2.38) can still be cast in the form of a variational principle

δP = δ

_

σ

s

(Γ

α

, ∇Γ

α

, . . .) dV = 0, (2.41)

where the time derivative symbol d/dt has been replaced by the variational

symbol δ. Since the corresponding Euler–Lagrange equations are shown to

be the stationary balance relations, it turns out that the stationary state is

characterized by an extremum of the entropy production, truly a minimum,

as it can be proved that the second variation is positive deﬁnite δ

2

P > 0.

It should also be realized that the minimum entropy principle is not an

extra law coming in complement of the classical balance equations of mass,

momentum, and energy, but nothing else than a reformulation of these laws

in a condensed form, just like in classical mechanics, Hamilton’s principle is

a reformulation of Newton’s equations.

The search for variational principles in continuum physics has been a sub-

ject of continuous and intense activity (Glansdorﬀ and Prigogine 1964, 1971;

Finlayson 1972; Lebon 1980). A wide spectrum of applications in macroscopic

physics, chemistry, engineering, ecology, and econophysics is discussed in

Sieniutycz and Farkas (2004). It should, however, be stressed that it is only in

exceptional cases that there exists a “true” variational principle for processes

that dissipate energy. Most of the principles that have been proposed refer

54 2 Classical Irreversible Thermodynamics

either to equilibrium situations, as the maximum entropy principle in equilib-

rium thermodynamics, the principle of virtual work in statics, the minimum

energy principle in elasticity, or to ideal reversible motions as the principle

of least action in rational mechanics or the minimum energy principle for

Eulerian ﬂuids.

2.6 Applications to Heat Conduction, Mass Transport,

and Fluid Flows

To better understand and illustrate the general theory, we shall deal with

some applications, like heat conduction in a rigid body and matter diﬀusion

involving no coupling of diﬀerent thermodynamic forces and ﬂuid ﬂow. More

complex processes involving coupling, like thermoelectricity, thermodiﬀusion

and diﬀusion through membranes are treated in Chap. 3. The selection of

these problems has been motivated by the desire to propose a pedagogical

approach and to cover situations frequently met in practical problems by

physicists, chemists and engineers. Chemistry will receive a special treatment

in Chap. 4 where we deal at length with chemical reactions and their coupling

with mass transport, a subject of utmost importance in biology. Despite its

success, CIT has been the subject of several limitations and criticisms, which

are discussed in Sect. 2.7 of the present chapter.

2.6.1 Heat Conduction in a Rigid Body

The problem consists in ﬁnding the temperature distribution in a rigid body

at rest, subject to arbitrary time-dependent boundary conditions on tem-

perature, or on the heat ﬂux. Depending on the geometry and the physical

properties of the system and on the nature of the boundary conditions, a

wide variety of situations may arise, some of them being submitted as prob-

lems at the end of the chapter. For the sake of pedagogy, we follow the same

presentation as in Sect. 2.4.

Step 1. State variable(s)

Here we may select indiﬀerently the speciﬁc internal energy u(r, t) or the

temperature ﬁeld T(r, t), which should be preferred in practical applications.

Step 2. Evolution equation

In absence of source term, the evolution equation (2.18) for u(r, t) is simply

ρ

du

dt

= −∇ q, (2.42)

here d/dt reduces to the partial time derivative ∂/∂t as v = 0. Equation (2.42)

contains two unknown quantities, the heat ﬂux q to be given by a constitutive

2.6 Applications to Heat Conduction, Mass Transport, and Fluid Flows 55

relation and the internal energy u(T) to be expressed by means of an equation

of state.

Step 3. Entropy production and second law

According to the second law, the rate of entropy production deﬁned by

σ

s

= ρ

ds

dt

+∇ J

s

≥ 0, (2.43)

is positive deﬁnite. The expression of ds/dt is obtained from Gibbs’ equation

ds

dt

= T

−1

du

dt

, (2.44)

where du/dt is given by the energy balance equation (2.42). Substituting

(2.44) in (2.43) results in

σ

s

= q ∇T

−1

+∇ (J

s

−T

−1

q). (2.45)

Since σ

s

represents the rate of entropy production inside the body, its ex-

pression cannot contain a ﬂux term like ∇ (J

s

−T

−1

q), which describes the

rate of exchange with the outside, as a consequence this term must be set

equal to zero so that

J

s

= T

−1

q, (2.46)

whereas (2.45) of σ

s

reduces to

σ

s

= q ∇T

−1

. (2.47)

This illustrates the general statement (2.20) that the entropy production is

a bilinear form in the force ∇T

−1

(the cause) and the ﬂux of energy q (the

eﬀect).

Step 4. Linear ﬂux–force relation

The simplest way to ensure that σ

s

≥ 0 is to assume a linear relationship

between the heat ﬂux and the temperature gradient; for isotropic media,

q = L

qq

(T)∇T

−1

, (2.48)

where L

qq

(T) is a scalar phenomenological coeﬃcient depending generally on

the temperature. Deﬁning the heat conductivity by λ(T) = L

qq

(T)/T

2

, the

ﬂux–force relation (2.48) takes the more familiar form

q = −λ∇T, (2.49)

which is nothing else than the Fourier’s law stating that the heat ﬂux is

proportional to the temperature gradient. We observe in passing that Curie’s

principle is satisﬁed as (2.49) is a relationship between ﬂux and force of the

same tensor character, namely vectors. In an anisotropic crystal, Fourier’s

relation reads as

q = −λ ∇T, (2.50)

where the heat conductivity λ is now a tensor of order 2.

56 2 Classical Irreversible Thermodynamics

Step 5. Restriction on the sign of the transport coeﬃcients

Substitution of (2.49) in (2.48) of the rate of entropy production yields

σ

s

=

1

λT

2

q q, (2.51)

and from the requirement that σ

s

≥ 0, it is inferred that λ ≥ 0. Roughly

speaking this means that in an isotropic medium, the heat ﬂux takes place

in a direction opposite to the temperature gradient; therefore, heat will ﬂow

spontaneously from high to low temperature, in agreement with our everyday

experience. In an anisotropic system, ﬂux and force will generally be oriented

in diﬀerent directions but the positiveness of tensor λ requires that the angle

between them cannot be smaller than π/2.

Step 6. Reciprocal relations

In the general case of an anisotropic medium, the ﬂux–force relation is the

Fourier’s law expressed in the form (2.50). According to Onsager’s recipro-

cal relations, the second-order tensor λ is symmetric so that, in Cartesian

coordinates,

λ

ij

= λ

ji

, (2.52)

a result found to be experimentally satisﬁed in crystals wherein, however,

spatial symmetry may impose further symmetry relations. For instance, in

crystals pertaining to the hexagonal or the tetragonal class, spatial symme-

try requires that the conductivity tensor is skew-symmetric. By combining

this result with the symmetry property (2.52), it turns out that the elements

λ

ij

(i ,= j) are zero; it follows that for these classes of crystal, the application

of a temperature gradient in the x-direction cannot produce a heat ﬂow in

the perpendicular y-direction. Very old experiences by Soret (1893) and Voigt

(1903) conﬁrmed this result, which is presented as one of the conﬁrmations

of the Onsager–Casimir’s reciprocal relations.

Step 7. The temperature equation

We now wish to calculate the temperature distribution in an isotropic rigid

solid as a function of time and space. The corresponding diﬀerential equation

is easily obtained by introducing Fourier’s law (2.49) in the energy balance

equation (2.42) and the result is

ρc

v

∂T

∂t

= ∇ (λ∇T), (2.53)

where use is made of the deﬁnition of the heat capacity c

v

= ∂u/∂T.

In the case of constant heat conductivity and heat capacity, and introduc-

ing the heat diﬀusivity deﬁned by χ = λ/ρc

v

, (2.53) reads as

∂T

∂t

= χ

∂

2

T

∂x

2

. (2.54)

Relation (2.53) is classiﬁed as a parabolic partial diﬀerential equation. In

Box 2.3 is presented the mathematical method of solution of this important

2.6 Applications to Heat Conduction, Mass Transport, and Fluid Flows 57

equation under given typical initial and boundary conditions for an inﬁnite

one-dimensional rod. It is the same kind of equation that governs matter

diﬀusion, as shown in Sect. 2.6.2.

Box 2.3 Method of Solution of the Heat Diﬀusion Equation

A convenient method to solve (2.54) is to work in the Fourier space (k, t)

with k designating the wave number. It is interesting to recall that Fourier

devised originally the transform bearing his name to solve the heat diﬀusion

equation. Let us write T(x, t) as a Fourier integral of the form

T(x, t) =

_

+∞

−∞

T(k, t) exp(ikx) dk, (2.3.1)

with T(k, t) the Fourier transform. The initial and boundary conditions

are assumed to be given by T(x, 0) = g(x), T(±∞, t) = 0; introduction of

(2.3.1) in (2.54) leads to the ordinary diﬀerential equation

dT(k, t)

dt

= −k

2

χT(k, t), (2.3.2)

whose solution is directly given by

T(k, t) = T(k, 0) exp(−k

2

χt), (2.3.3)

where T(k, 0) is the Fourier transform of the initial temperature proﬁle

T(k, 0) =

1

2π

_

+∞

−∞

g(x

) exp(−ikx

) dx

. (2.3.4)

Substitution of (2.3.3) and (2.3.4) in (2.3.1) yields T(x, t) in terms of the

initial distribution g(x):

T(x, t) =

1

2π

_

+∞

−∞

dk

_

+∞

−∞

g(x

) exp(−k

2

χt) exp[ik(x −x

)] dx

. (2.3.5)

By carrying the integration with respect to k, we obtain the ﬁnal solution

in the form

T(x, t) =

1

(4πχt)

1/2

_

+∞

−∞

g(x

) exp[−(x −x

)

2

/4χt] dx

. (2.3.6)

When the initial temperature dependence corresponds to a local heating at

one particular point x

0

of the solid, namely

g(x) = g

0

δ(x −x

0

), (2.3.7)

where g

0

is an arbitrary constant proportional to the energy input at the

initial time, and δ(x −x

0

) the Dirac function, (2.3.6) will be given by

T(x, t) =

g

0

(4πχt)

1/2

exp[−(x −x

0

)

2

/4χt]. (2.3.8)

58 2 Classical Irreversible Thermodynamics

We deduce from (2.3.8) that the temperature diﬀuses in the whole rod from

the point x

0

over a distance proportional to (χt)

1/2

; as a consequence, large

values of the thermal diﬀusivity imply a rapid diﬀusion of temperature. Solu-

tion (2.3.8) exhibits also the property that, after application of a temperature

disturbance at a given point of the system, it will be experienced instanta-

neously everywhere in the whole body. Such a result is in contradiction with

the principle of causality, which demands that response will be felt after the

application of a cause. In the above example, cause and eﬀect occur simul-

taneously. This property is typical of classical irreversible thermodynamics

but is not conceptually acceptable. This failure was one of the main motiva-

tions to propose another thermodynamic formalism currently known under

the name of Extended Irreversible Thermodynamics (see Chap. 7).

Box 2.4 Stationary States and Minimum Entropy Production

Principle

In the case of thermal conduction described by the phenomenological equa-

tion q = L

qq

∇T

−1

, the total entropy production P in the system, say a

one-dimensional rod of length l, is according to (2.47),

P =

_

l

0

σ

s

dx =

_

l

0

q(∂T

−1

/∂x) dx =

_

l

0

L

qq

(∂T

−1

/∂x)

2

dx. (2.4.1)

For a constant phenomenological coeﬃcient L

qq

, its time derivative is

dP

dt

= 2

_

l

0

L

qq

∂T

−1

∂x

∂

∂t

∂T

−1

∂x

dx. (2.4.2)

After integration by parts and imposing time-independent boundary con-

ditions, one has

dP

dt

= −2

_

l

0

∂T

−1

∂t

∂

∂x

_

L

qq

∂

∂x

T

−1

_

dx = 2

_

l

0

ρ

∂T

−1

∂t

∂u

∂t

dx, (2.4.3)

where use is made of the energy balance equation (2.42). Since ∂u/∂t =

c

v

∂T/∂t with c

v

> 0, it is ﬁnally found that

dP

dt

= −2

_

l

0

ρc

v

T

2

_

∂T

∂t

_

2

dx ≤ 0 (2.4.4)

is negative outside the stationary state and zero in the stationary state for

which ∂T/∂t = 0.

It is directly recognized that the total entropy production P is a Lya-

pounov function, as P > 0 together with dP/dt ≤ 0, and this ensures

that the stationary state is stable. It is, however, important to realize that

the phenomenological coeﬃcient L

qq

is related to the heat conductivity by

L

qq

= λT

2

and therefore the validity of the minimum entropy production

2.6 Applications to Heat Conduction, Mass Transport, and Fluid Flows 59

principle is conditioned by the condition that λ varies like T

−2

; the princi-

ple is therefore not applicable to systems with a constant heat conductivity

or with an arbitrary dependence of λ with respect to the temperature. This

is the reason why the criterion of minimum entropy production remains an

exception.

2.6.2 Matter Diﬀusion Under Isothermal and Isobaric

Conditions

In this section, we brieﬂy discuss the problem of isothermal and isobaric

diﬀusion of two non-viscous isotropic non-reacting ﬂuids, in the absence of

external forces. There is no diﬃculty to generalize the forthcoming consid-

erations and results to the more general case of a n-component mixture of

viscous ﬂuids.

Let ρ

1

and ρ

2

denote the densities (mass per unit volume) and c

1

(=

ρ

1

/ρ), c

2

(= ρ

2

/ρ) the mass fractions of the two constituents (c

1

+ c

2

= 1),

ρ = ρ

1

+ρ

2

is the total mass density. Designating by v

k

(r, t) the local macro-

scopic velocity of substance k, the centre of mass or barycentric velocity is

given by

ρv = ρ

1

v

1

+ρ

2

v

2

. (2.55)

The diﬀusion ﬂux J

k

of substance k with respect to the centre of mass is

deﬁned by

J

k

= ρ

k

(v

k

−v), (2.56)

and is expressed in kg m

−2

s

−1

; in some circumstances, it may be convenient

to replace in (2.56) the barycentric velocity by an arbitrary reference velocity

(de Groot and Mazur 1962). In virtue of the deﬁnition of the barycentric

velocity, it is directly seen that J

1

+J

2

= 0.

The analysis carried out in Sect. 2.6.1 for heat transport is easily repeated

for matter diﬀusion. In the present problem, the Gibbs’ equation takes the

form

T

ds

dt

=

du

dt

+p

dv

dt

−(¯ µ

1

− ¯ µ

2

)

dc

1

dt

, (2.57)

wherein ¯ µ

1

and ¯ µ

2

are the respective chemical potentials and where use has

been made of dc

2

= −dc

1

. Combining (2.57) with the balance equations

(2.16)–(2.18) written without source terms, it is easily shown (see Prob-

lem 2.6) that the entropy production is given by

Tσ

s

= −J

1

∇(¯ µ

1

− ¯ µ

2

) ≥ 0. (2.58)

Within the hypothesis of linear ﬂux–force relations, one obtains

J

1

= −L∇(¯ µ

1

− ¯ µ

2

) (L > 0), (2.59)

60 2 Classical Irreversible Thermodynamics

or, in virtue of Gibbs–Duhem’s relation c

1

d¯ µ

1

+c

2

d¯ µ

2

= c

1

∇¯ µ

1

+c

2

∇¯ µ

2

= 0,

J

1

= −

L

c

2

∇¯ µ

1

= −

L(∂¯ µ

1

/∂c

1

)

T,p

c

2

∇c

1

, (2.60)

wherein (∂¯ µ

1

/∂c

1

)

T,p

≥ 0 to satisfy the requirement of stability of equilib-

rium. After identifying the coeﬃcient of ∇c

1

with ρD, where D (in m

2

s

−1

)

is the positive diﬀusion coeﬃcient, one ﬁnds back the celebrated Fick’s law

J

1

= −ρD∇c

1

(D > 0). (2.61)

Elimination of J

1

between Fick’s law and the balance equation ρ dc

1

/dt =

−∇ J

1

leads to a parabolic diﬀerential equation in c

1

(r, t) similar to (2.54).

In the particular case of diﬀusion of N material particles located at a point

r = r

0

at time t

0

, the distribution of the mass fraction c

1

(r, t) obeys the

same law as (2.3.8), namely

c

1

(r, t) =

N

8(πDt)

3/2

exp[−(r −r

0

) (r −r

0

)/(4Dt)]. (2.62)

The characteristic displacement length is now given by l ∝

√

Dt, and this

result constitutes the main characteristic of diﬀusion phenomena, namely

that the mean distance of diﬀusion is proportional to the square root of the

time. Note that it is also usual to express Fick’s law in terms of the molar

concentration rather than the mass fraction; in such a case, Fick’s law takes

the form (2.2) instead of (2.61).

2.6.3 Hydrodynamics

As a further application, we will show that CIT provides a general frame-

work for the macroscopic description of hydrodynamics. The problem that

we wish to solve is the establishment of the complete set of partial diﬀeren-

tial equations governing the motion of a ﬂuid. To be explicit, let us consider

the laminar motion (turbulence is excluded) of a one-constituent isotropic,

compressible and viscous ﬂuid in presence of a temperature gradient, without

internal energy supply.

The set of basic variables giving a complete knowledge of the system in

space and time are the mass density ρ(r, t), the velocity v(r, t) and the

temperature T(r, t) ﬁelds.

These variables obey the classical balance equations of mass, momentum,

and energy already given in Sect. 2.4 but recalled here for the sake of clarity:

ρ

dρ

−1

dt

= ∇ v, (2.63)

ρ

dv

dt

= −∇ P+ρF, (2.64)

ρ

du

dt

= −∇ q −P : V. (2.65)

2.6 Applications to Heat Conduction, Mass Transport, and Fluid Flows 61

The quantity V is the symmetric part of the velocity gradient tensor; in

Cartesian coordinates, V

ij

=

1

2

(∂v

i

/∂x

j

+ ∂v

j

/∂x

i

). In ﬂuid mechanics, the

symmetric pressure tensor P is usually split into a reversible hydrostatic

pressure pI and an irreversible viscous pressure P

v

in order that P = pI +

P

v

with I the identity tensor. The symmetric viscous pressure tensor P

v

may further be decomposed into a bulk part p

v

(=

1

3

tr P

v

) and a traceless

deviatoric part

0

P

v

so that P = p

v

I+

0

P

v

. The possibility and physical meaning

of an antisymmetric contribution to P

v

will be analysed in Problem 2.9.

Gathering all these results, the pressure tensor can be written as

P = (p +p

v

)I +

0

P

v

. (2.66)

Instead of the pressure tensor P, some authors prefer to use the stress tensor

σ, which is equal to minus the pressure tensor. All the results derived above

and the forthcoming remain valid by working with the stress tensor. The

evolution equations (2.63)–(2.65) constitute a set of ﬁve scalar relations with

16 unknown quantities namely, ρ, v, p, p

v

,

0

P

v

, q, u, and T. It is the aim

of the classical theory of irreversible processes to provide the eleven missing

equations. As usual, we start from Gibbs’ relation

ds

dt

= T

−1

du

dt

−pT

−1

dρ

−1

dt

, (2.67)

wherein du/dt and dρ

−1

/dt will be replaced, respectively, by expressions

(2.65) and (2.63). This yields the following balance equation for the speciﬁc

entropy s:

ρ

ds

dt

= −∇

q

T

+q ∇T

−1

−T

−1

p

v

∇ v −T

−1

0

P

v

:

0

V, (2.68)

where

0

V is the traceless part of tensor V. From (2.68), it is inferred that the

expressions of the entropy ﬂux and the entropy production are given by

J

s

=

1

T

q, (2.69)

σ

s

= q ∇T

−1

−T

−1

p

v

∇ v −T

−1

0

P

v

:

0

V. (2.70)

Expression (2.70) is bilinear in the ﬂuxes q, p

v

,

0

P

v

, and the forces ∇T

−1

,

T

−1

(∇ v), T

−1

0

V; by assuming linear relations between them and invoking

Curie’s law, one obtains the following set of phenomenological relations, also

called the transport equations:

q = L

qq

∇T

−1

≡ −λ∇T, (2.71)

p

v

= −l

vv

T

−1

∇ v ≡ −ζ∇ v, (2.72)

0

P

v

= −L

vv

T

−1

0

V ≡ −2η

0

V. (2.73)

62 2 Classical Irreversible Thermodynamics

We recognize in (2.71) the Fourier’s law by identifying the phenomenolog-

ical coeﬃcient L

qq

/T

2

with the heat conductivity λ; (2.72) is the Stokes’

relation if l

vv

/T is identiﬁed with the bulk viscosity ζ, and ﬁnally (2.73) is

the Newton’s law of hydrodynamics when L

vv

/T is put equal to 2η, with η

the dynamic shear viscosity. It follows from (2.71)–(2.73) that the ﬂuid will be

set instantaneously in motion as soon as it is submitted to a force, in contrast

with elastic solids, which may stay at rest even when subject to stresses. At

equilibrium, both members of equations (2.71)–(2.73) vanish identically as it

should. The phenomenological coeﬃcients λ, ζ, and η depend generally on ρ

and T but in most practical situations the dependence with respect to ρ is

negligible. In the case of incompressible ﬂuids for which ∇ v = 0, the viscous

pressure p

v

is zero and the bulk viscosity does not play any role. However,

in compressible ﬂuids like dense gases and bubbly liquids, the bulk viscosity

is by no means negligible. The bulk viscosity vanishes in the case of perfect

gases; in real gases, it is far from being negligible and the ratio ζ/η between

bulk and shear viscosity may even be of the order of hundred.

After substitution of the transport equations (2.71)–(2.73) in (2.70) of the

entropy production, one obtains

σ

s

=

λ

T

2

(∇T)

2

+

ζ

T

(∇ v)

2

+

2η

T

(

0

V :

0

V) ≥ 0. (2.74)

Positiveness of σ

s

requires that

λ > 0, ζ > 0, η > 0. (2.75)

Negative values of these transport coeﬃcients will therefore be in contradic-

tion with the second law of thermodynamics.

By combining the nine transport equations (2.71)–(2.73) with the ﬁve

evolution equations (2.63)–(2.65), one is faced with a set of 14 equations

with 16 unknowns. The missing relations are provided by Gibbs’ equation

(2.67) from which are supplemented the two equations of state T = T(u, ρ),

p = p(u, ρ), or solving with respect to u,

u = u(T, ρ), p = p(T, ρ). (2.76)

These equations, together with the evolution and transport relations describe

completely the behaviour of the one-component isotropic ﬂuid after that

boundary and initial conditions have been speciﬁed. The state equations

(2.76) have a status completely diﬀerent from the phenomenological equa-

tions (2.71)–(2.73) because, in contrast with the latter, they do not vanish

identically at equilibrium. By substitution of the phenomenological laws of

Stokes (2.72) and Newton (2.73) in the momentum balance equation (2.64),

one ﬁnds back the well-known Navier–Stokes’ equation when it is assumed

that the viscosity coeﬃcients are constant:

ρ

dv

dt

= −∇p + 2η∇

2

v +

_

2

3

η +ζ

_

∇(∇ v) +ρF. (2.77)

2.7 Limitations of the Classical Theory of Irreversible Thermodynamics 63

In the case of an isothermal and incompressible ﬂuid, which is the most

customary situation met in hydrodynamics, the behaviour of the ﬂuid is com-

pletely described by (2.77) and ∇ v = 0. For more general situations like

multi-component mixtures with diﬀusion, the equations become much more

complicated but the way to derive them follows the same systematic proce-

dure as above. It should, however, be stressed that the results of this section

are only applicable to the study of laminar ﬂows of some particular class of

ﬂuids, namely these described by the linear ﬂux–force relations of Fourier,

Stokes, and Newton. The above formalism is not adequate for non-Newtonian

ﬂuids or polymeric solutions and cannot cope with turbulent motions. We

note ﬁnally that the basic relations obtained so far, like for instance the

Navier–Stokes’ equation (2.77), are parabolic partial diﬀerential equations,

which means that disturbances will propagate through the ﬂuid with an in-

ﬁnite velocity.

2.7 Limitations of the Classical Theory

of Irreversible Thermodynamics

Despite its numerous successes, CIT has raised several questions and some

shortcomings have been pointed out. Let us comment brieﬂy about the most

frequent criticisms:

1. The theory is not applicable to irreversible processes described by non-

linear phenomenological equations. Many actual processes, like chemical

reactions or non-Newtonian ﬂows, are indeed characterized by non-linear

phenomenological relations and are therefore outside the scope of classical

irreversible thermodynamics. It should, however, be noted that linearity

between ﬂuxes and forces does not require that the phenomenological co-

eﬃcients, like the heat conductivity and the viscosity coeﬃcients, are con-

stant; they may indeed depend in particular on the temperature and the

pressure with the consequence that the corresponding evolution equations,

like Navier–Stokes’ equation, may be non-linear.

2. The cornerstone of the theory is the local equilibrium hypothesis, accord-

ingly the thermodynamic state variables are the same as in equilibrium

and Gibbs’ equation, which plays a fundamental role in CIT, remains lo-

cally valid, i.e. at each position in space and each instant of time. The

advantage of this assumption is that entropy has a precise meaning, even

in non-equilibrium states, and that it leads to an explicit expression for

the entropy production from which are inferred appropriate forms for the

constitutive relations. However, it is conceivable that other variables, not

found at equilibrium, are able to inﬂuence the process. This is, for instance,

the case of polymers of long molecular chains in which conﬁguration in-

ﬂuences considerably their behaviour. Other examples are superﬂuids and

superconductors whose peculiar properties ask for the introduction of extra

64 2 Classical Irreversible Thermodynamics

variables. As the local equilibrium hypothesis implies large time and space

scales, CIT is not appropriate for describing high-frequency phenomena

as ultrasound propagation or nuclear collisions and short-wavelength sys-

tems and processes, like nano-structures, shock waves or light and neutron

scattering.

Another important point to notice is that, when referred to general-

ized hydrodynamics, the phenomenological coeﬃcients are frequency and

wavelength dependent. This is at variance with the local equilibrium hy-

pothesis, implying that these coeﬃcients are frequency and wavelength

independent.

3. By introduction of the Fourier’s law with constant heat conductivity in

the energy balance, one obtains an equation for the temperature, which

is of the diﬀusion type; see for instance (2.54). From the mathematical

point of view, this is a parabolic partial diﬀerential equation, which im-

plies that after application of a disturbance, the latter will propagate at

inﬁnite velocity across the body so that it will be felt instantaneously and

everywhere in the whole body. The same remark is applicable to the other

equations of the theory and in particular to Navier–Stokes’ equation. The

problem of propagation of signals with an inﬁnite velocity is conceptually

a major inconvenient, which is not acceptable from a purely physical point

of view. However it is not a tragedy in most practical situations because

their characteristic time is much longer than the transit time of the signals

transporting the thermodynamic information.

4. Another important subject of controversy concerns the validity of the

Onsager–Casimir’s reciprocal relations, which were derived from the con-

dition of microscopic reversibility. Although the reciprocity property was

originally shown by Onsager to be applicable at the microscopic level for

particles in situations close to equilibrium, it is current to extrapolate

this result at the macroscopic level for continuum media driven far from

equilibrium. There is clearly no theoretical argument supporting such an

extrapolation. Onsager’s demonstration itself has been the subject of acrid

criticisms because it is based on an important assumption with regard to

the regression of the ﬂuctuating thermodynamic variables. Indeed, it lies

on the hypothesis that the rate of decay of the ﬂuctuations takes place

according to the same law as the macroscopic linear ﬂux–force relations.

Moreover, in the light of the demonstration given by Onsager, the thermo-

dynamic ﬂuxes are deﬁned as time derivatives of extensive thermodynamic

variables and the forces are the derivatives of the entropy with respect to

the same state variables. Such conditions are certainly not respected by

thermodynamic ﬂuxes like the heat ﬂux or the pressure tensor or ther-

modynamic forces as the temperature gradient or the velocity gradient

tensor.

5. As shown explicitly in Box 2.2, the Curie law stating that in isotropic ma-

terials, ﬂuxes couple only with forces of the same tensorial order, is directly

derivable from algebra. It is also important to realize that this exclusion

2.8 Problems 65

law is not valid outside the linear regime even for isotropic bodies. As

observed by Truesdell (1984), these results were well known even before

Curie was born. Moreover, when one goes back to Curie’s original papers

on symmetry considerations, there is nothing, even vaguely mentioned,

which is reminding of the law bearing his name in classical irreversible

thermodynamics. In anisotropic systems, the coupling between terms of

diﬀerent tensorial order, as for instance chemical reactions (scalar) and

mass transport (vectorial), may provide very important eﬀects, as shown

in Chap. 4.

6. Another point of debate is the question of how the ﬂuxes and forces are

to be selected from the expression of the entropy production. It is gener-

ally admitted that the thermodynamic ﬂuxes are the physical ﬂuxes that

appear in the balance laws of mass, momentum, and energy, and that the

forces are the conjugated terms in the bilinear expression of the entropy

production. Another suggestion is to identify the forces as the gradient

of intensive variables, as temperature and velocity, and to select as ﬂuxes

the conjugate terms in the entropy production. Unfortunately, such deﬁn-

itions are not applicable to chemical reactions. Another school of thought

claims that it is indiﬀerent how the ﬂuxes and forces are selected, at the

condition that non-singular linear transformations leave invariant the en-

tropy production expression. However, as pointed out by Coleman and

Truesdell (1960), one should be careful in the identiﬁcation of ﬂuxes and

forces, as a wrong selection could destroy the reciprocal property of the

phenomenological coeﬃcients. Clearly, the controversy about how ﬂuxes

and forces are chosen is not closed but, in our opinion, this is not the more

fundamental question about CIT, all the more as the discussions about

this point have been sterile.

2.8 Problems

2.1. Heat transport and entropy production. Two heat reservoirs at tempera-

ture T

1

and T

2

are connected by a rigid heat-conducting one-dimensional rod.

Find the longitudinal temperature proﬁles in the rod if the heat conductivity

is (a) constant and (b) proportional to T

α

, where α is a constant. Calculate

the entropy produced along the rod per unit time.

2.2. Fourier’s and Newton’s laws. Consider a rigid cylindrical heat-conducting

rod. The heat transfer along the rod is described by means of Fourier’s

law (q = −λ∇T), whereas the heat exchange with the environment is de-

scribed by Newton’s cooling law, expressing that the heat ﬂow is propor-

tional to the temperature diﬀerence between the rod and the environment

(q n = −α(T −T

env

)), where n is the unit vector normal to the boundaries.

(a) Derive the equation describing the temperature distribution along the

66 2 Classical Irreversible Thermodynamics

rod. (b) Determine the temperature proﬁle along a very long rod if its hotter

boundary is kept at T

1

and the environment’s temperature T

env

is constant.

2.3. Temperature proﬁle. A cylindrical tube of radius R and length L is im-

mersed in an environment at temperature T

0

. A ﬂuid ﬂows inside the tube at

a ﬂow rate Q; the temperature of the ﬂuid when it enters the tube is T

1

. Find

the temperature T

2

of the outgoing ﬂuid at the end of the tube as a function

of the ﬂow rate Q in terms of its speciﬁc heat per unit mass, its density and

its thermal conductivity. The heat transfer coeﬃcient across the lateral walls

of the tube is a.

2.4. Radiative heat transfer. Assume that heat exchange between a spher-

ical body of radius R and its environment occurs through electromagnetic

radiation and obeys Stefan–Boltzmann’s law. Accordingly, the quantity of

heat emitted per unit area and unit time by a body at temperature T is

q = εσT

4

, where σ is the Stefan–Boltzmann constant and ε the emissivity

of the system. (a) Establish the diﬀerential equation describing the cooling

of the sphere when its surface is at temperature T and the environment at

temperature T

env

. (b) Find the corresponding entropy production per unit

time. (c) Solve the problem when the diﬀerence T −T

env

is small compared

to T

env

. (d) How will the solution be modiﬁed if the radius of the sphere is

10 times larger?

2.5. The age of the Earth. A semi-inﬁnite rigid solid of heat diﬀusivity χ,

initially at temperature T

10

, is suddenly put into contact with another semi-

inﬁnite solid at ﬁxed temperature T

2

. Assume that heat exchange is described

by Fourier’s law. (a) Show that the temperature proﬁle as a function of time

and position is

θ(x, t) =

θ

0

√

πχt

_

x

−∞

dx

exp

_

−

(x −x

)

2

4χt

_

,

where θ = T(x, t) − T

2

and θ

0

≡ T

10

− T

2

. (b) Verify that the temperature

gradient at the surface is

_

∂θ

∂x

_

x=0

= −

θ

0

√

πχt

.

(c) At the end of the nineteenth century, lord Kelvin evaluated the age of

the Earth from an analysis of the terrestrial temperature gradient near the

surface of the Earth, found to be equal to −3 10

−2◦

Cm

−1

. By taking

χ ≈ 310

5

m

2

s

−1

and θ

0

≈ 3, 800

◦

C, estimate the age of the Earth according

to this model. The result found is much less than the actual estimation of

4.5 billion years (radioactivity, which was not known at Kelvin’s time, will

considerably inﬂuence the result).

2.6. Matter diﬀusion. Derive the expressions of the entropy ﬂux and the

entropy production for the problem of isothermal and isobaric diﬀusion in a

binary mixture of non-viscous ﬂuids, in absence of external body forces (see

Sect. 2.6.2).

2.8 Problems 67

2.7. Concentration proﬁle. A perfectly absorbing sphere of radius R is im-

mersed in a solvent–solute solution with homogeneous concentration c

0

. The

diﬀusion coeﬃcient of the solute is D. Since the sphere is perfectly absorbing,

the concentration of solute on its surface is always zero. (a) Determine the

concentration proﬁle as a function of the radial distance. (b) Show that the

amount of solute crossing the surface of the sphere per unit time is given

by I = 4πDRc

0

. Hint: In spherical symmetry, the diﬀusion equation has the

form

∂c

∂t

=

D

r

2

∂

∂r

_

r

2

∂c

∂r

_

.

2.8. The Einstein relation. A dilute suspension of small particles in a viscous

ﬂuid at homogeneous temperature T is under the action of the gravitational

ﬁeld. The friction coeﬃcient of the particles with respect to the ﬂuid is a (a =

6πηr for spherical particles of radius r in a solvent with viscosity η). Owing to

gravity, the particles have a sedimentation velocity v

sed

= m

g/a, with m

the

mass of one particle minus the mass of the ﬂuid displaced by it (Archimede’s

principle); the corresponding sedimentation ﬂux is J

sed

= nv

sed

, with n the

number of particles per unit volume. Against the sedimentation ﬂux is acting

a diﬀusion ﬂux J

dif

= −D∂n/∂z, D being the diﬀusion coeﬃcient. (a) Find

the vertical proﬁle of n(z) in equilibrium. (b) Compare this expression with

Boltzmann’s result

n(z) = n(0) exp[−(m

gz/k

B

T)],

where k

B

is the Boltzmann constant, and demonstrate Einstein’s relation

D = k

B

T/a.

2.9. Micropolar ﬂuids. In some ﬂuids (composed of elongated particles or

rough spheres) the pressure tensor is non-symmetric. Its antisymmetric part

is related to the rate of variation of an intrinsic angular momentum, and

therefore, it contributes to the balance equation of angular momentum. The

antisymmetric part of the tensor is usually related to an axial vector P

va

,

whose components are deﬁned as P

va

1

= P

va

23

, P

va

2

= P

va

31

, P

va

3

= P

va

12

. The

balance equation for the internal angular momentum is

ρj ˙ ω +∇ Q = −2P

va

,

with j the microinertia per unit mass of the ﬂuid, w the angular velocity, and

Q the ﬂux of the intrinsic angular momentum, which is usually neglected.

(a) Prove that the corresponding entropy production is

σ

s

= −T

−1

p

v

∇ v −T

−1

0

P

vs

: (

0

∇v)

s

−T

−1

P

va

(∇v −2ω) +q ∇T

−1

,

where

0

P

vs

is the symmetric part of P

v

. Hint: Note that ∂s/∂ω = −ρjT

−1

ω.

(b) Show that the constitutive equation for P

va

may be given by

68 2 Classical Irreversible Thermodynamics

P

va

= −η

r

(∇v −2ω),

where η

r

is the so-called rotational viscosity; explain why there is no coupling

between P

va

and q (see, for instance, Snider and Lewchuk 1967; Rub´ı and

Casas-V´azquez 1980).

2.10. Flux–force relations. It is known that by writing a linear relation of the

form J

α

=

β

L

αβ

X

β

between n-independent ﬂuxes J

α

(α = 1, . . . , n) and

n-independent thermodynamic forces X

β

, the matrix L

αβ

of phenomenologi-

cal coeﬃcients is symmetric, according to Onsager. (a) Show that the matrix

L

αβ

is still symmetric when not all the ﬂuxes are independent but one of

them is a linear combination of the other ones, i.e.

n

α=1

a

α

J

α

= 0. (b) Does

this conclusion remain fulﬁlled when instead of a linear relation between the

ﬂuxes, there exists a linear relation

n

α=1

a

α

X

α

= 0 between the forces?

2.11. Minimum entropy production principle. Consider a one-dimensional

rigid heat conductor with ﬁxed temperatures at its ends. Show that the

Euler–Lagrange equation corresponding to the principle of minimum entropy

production

δP(T

−1

) = δ

_

q(∂T

−1

/∂x) dx = 0

is the steady energy balance ∂q/∂x = 0 at the condition to write Fourier’s

law in the form q = L∂T

−1

/∂x with L = constant. Show that δ

2

P ≥ 0; this

result expresses that the Euler–Lagrange equation is a necessary condition

for P to be a minimum. Is the principle still fulﬁlled when Fourier’s law is

written in the more classical form q = −λ∂T/∂x with λ = constant?

2.12. Variational principles. Prigogine’s minimum entropy production is not

the only variational principle in linear non-equilibrium thermodynamics.

Other principles have been proposed, amongst others, by Onsager (1931) and

Gyarmati (1970). For instance, consider the entropy production P(X, J) ≡

i

X

i

J

i

and the dissipation functions

Φ(J, J) ≡

1

2

i,j

R

ij

J

i

J

j

and Ψ(X, X) ≡

1

2

i,j

L

ij

X

i

X

j

,

X being the thermodynamic forces and J the conjugated ﬂuxes. Show that if

the matrices L

ij

, R

ij

, and L

ij

are constant and symmetric, the two following

variational principles are valid (1) for prescribed forces X and varying ﬂuxes

J, P(X, J) − Φ(J, J) is maximum implies that X

k

=

i

R

ki

J

i

and (2) for

prescribed ﬂuxes J and varying forces X, P(X, J) − Ψ(X, X) is maximum

implies that J

k

=

i

L

ki

X

i

. (The reader interested in these topics is referred

to Ichiyanagi 1994.)

Chapter 3

Coupled Transport Phenomena

Thermoelectricity, Thermodiﬀusion, Membranes

The analysis of coupled processes is one of the most outstanding aspects of

the classical theory of non-equilibrium thermodynamics. To emphasize this

feature, this chapter is dedicated to the study of coupled transport processes,

involving two ﬂuxes and two thermodynamic forces. Emphasis is put on some

practical applications in thermoelectricity (coupling of heat and electric-

ity), thermodiﬀusion (coupling of heat and mass motion) and transport of

matter through membranes. In the precedent chapter, we have already out-

lined, in general terms, the importance of the Onsager’s reciprocal relations

ruling the coeﬃcients that describe such couplings. Here, we will discuss di-

rect applications.

Coupled transport processes ﬁnd a natural theoretical framework in non-

equilibrium thermodynamics. However, they were discovered well before the

developments of these theories during the early nineteenth century. Indeed,

thermoelectric eﬀects were observed by Seebeck and Peltier, respectively, in

1821 and 1835, even earlier than the quantitative formulation of Ohm’s law in

1855. The ﬁrst results on thermodiﬀusion were obtained by Ludwig in 1856,

Dufour’s eﬀect was discovered in 1872, and Soret carried out his ﬁrst system-

atic researches in 1879. Although these topics are rather old and classical,

from the practical point of view they have entered recently in a completely

new perspective, because of the recent developments in materials sciences,

high-power lasers and renewed interest in optimization of energy generation.

We will outline here these new features.

Transport properties play a central role in materials sciences. For instance,

a problem of acute interest is to obtain superconductor materials with high

critical temperatures. Another domain of intense activity is the optimal ap-

plication of thermoelectricity to convert waste heat to usable electricity or to

develop cooling devices; this requires materials with high Seebeck coeﬃcient,

high electrical conductivity and low thermal conductivity. This is not an easy

task because usually a high electrical conductivity implies also a high ther-

mal conductivity. Recently, research on thermoelectric materials has known

an extraordinary burst: it is frequent to ﬁnd semiconductors and superlattice

69

70 3 Coupled Transport Phenomena

devices whose ﬁgure of merit (a combination of Seebeck coeﬃcient, electrical

and thermal conductivities directly related to the eﬃciency of energy con-

version) has increased by more than one order of magnitude since the end of

the twentieth century (Mahan et al. 1997). This was in particular achieved

in superlattices in which one dimension is smaller than the mean free path of

phonons but larger than that of electrons. More generally, it is desirable to

combine a macroscopic approach of transport phenomena, to design the most

eﬃcient devices, with a microscopic understanding of the basic phenomena.

In parallel with thermoelectricity, practical applications of thermodiﬀusion

have also known a renewal of interest during the last decades with the possi-

bility of producing much localized heating of ﬂuids by means of high-power

and extremely focused lasers. High temperature gradients will promote the

separation of diﬀerent constituents in mixtures.

Finally, since the problem of transport across membranes remains a subject

of constant and wide interest, mainly in life sciences, we have reserved one

section of this chapter to its analysis.

3.1 Electrical Conduction

The results presented in Chap. 2 are directly applicable to the problem of

motion of electric charges in a conductor. First, we brieﬂy study the ﬂow

of electrical carriers in a rigid conductor at rest and in absence of thermal

eﬀects. The basic variables are (1) z, the electrical charge of the carrier, say

electrons, per unit mass (z should not be confused with the charge z

e

per

unit mass of the electrons to which it is related by z = c

e

z

e

, where c

e

is the

mass fraction of electrons) and (2) u the internal energy per unit mass. The

evolution equations are the law of conservation of charges and the internal

energy balance, i.e.

ρ

dz

dt

= −∇ i , (3.1)

ρ

du

dt

= E i , (3.2)

where E is the electric ﬁeld and i the density of conductive current (current

intensity per unit transversal area); in whole generality, the latter is related

to the density I of total current I by I = i +ρzv, where ρzv is the convective

current. In the present problem, one has v = 0 so that I = i while d/dt will

reduce to the partial time derivative.

The Gibbs’ equation (2.4) simpliﬁes as T ds = du − µ

e

dc

e

, with µ

e

the

chemical potential of the electrons. In terms of the time derivative and the

charge density z, Gibbs’ relation takes the form

T

ds

dt

=

du

dt

−

µ

e

z

e

dz

dt

. (3.3)

3.1 Electrical Conduction 71

Substitution of (3.1) and (3.2) in (3.3) yields

ρ

ds

dt

= ∇

_

µ

e

Tz

e

i

_

+

1

T

_

E −∇

µ

e

z

e

_

i , (3.4)

from which it is deduced that the entropy ﬂux and the entropy production

are given, respectively, by

J

s

=

µ

e

Tz

e

i , (3.5)

Tσ

s

=

_

E −∇

µ

e

z

e

_

i ≥ 0. (3.6)

The last result suggests writing the following linear ﬂux–force relation

E −∇

µ

e

z

e

= ri , (3.7)

or, for negligible values of ∇(µ

e

/z

e

),

E = ri , (3.8)

which is Ohm’s law with r the electrical resistivity. After introducing Ohm’s

law in (3.6), one obtains Tσ

s

= ri

2

, which represents Joule’s heating and

from which is inferred that the electrical resistivity is a positive quantity.

In the presence of an external magnetic ﬁeld H, interesting coupled eﬀects

appear and the resistivity is no longer a scalar, but a second-order tensor

because of the anisotropy introduced by H. In this case, the resistivity de-

pends on the magnetic ﬁeld and the Onsager–Casimir’s reciprocity relations

state that r

ij

(H) = r

ji

(−H). If the external magnetic ﬁeld is applied in

the z-direction with all currents and gradients parallel to the x–y plane, the

resistivity tensor r will take the form (Problem 3.2)

r =

⎛

⎜

⎝

r

11

(H) r

12

(H) 0

−r

12

(H) r

22

(H) 0

0 0 r

33

(H)

⎞

⎟

⎠, (3.9)

where H is the component in the z-direction. It follows that an electric ﬁeld

in the x-direction will produce a current not only in the parallel direction but

also in the normal y-direction, a phenomenon known as the isothermal Hall

eﬀect. This eﬀect is easily understandable, as the external magnetic ﬁeld will

deﬂect the charge carriers along a curved trajectory. The discovery, at the

beginning of the 1980s, of subtle phenomena such as quantum Hall eﬀect,

boosted an enormous impetus on this topic. The dependence of the diagonal

components on the magnetic ﬁeld, known as magneto-resistance, has led to

important applications in computer technology. These topics, however, are

beyond the scope of purely macroscopic considerations.

72 3 Coupled Transport Phenomena

3.2 Thermoelectric Eﬀects

The coupling between temperature gradient and electrical potential gradient

gives rise to the thermoelectric eﬀects of Seebeck, Peltier, and Thomson. Here

we present them in the framework of classical irreversible thermodynamics,

outline their main physical features and, as a practical application, we will

determine the eﬃciency of thermoelectric generators.

3.2.1 Phenomenological Laws

In presence of both thermal and electrical eﬀects, the law of conservation

of charge (3.1) remains unchanged while the energy balance equation (3.2)

contains an additional term and reads as

ρ

du

dt

= −∇ q +E i . (3.10)

Combining (3.1) and (3.10) with the Gibbs’ equation (3.3), it is a simple

exercise (see Problem 3.1) to prove that the entropy ﬂux and the rate of

dissipation per unit volume Tσ

s

are, respectively, given by

J

s

=

q

T

−

µ

e

Tz

e

i , (3.11)

Tσ

s

= −J

s

∇T +

_

E −∇

_

µ

e

z

e

__

i . (3.12)

Restricting to isotropic media, the corresponding phenomenological equations

are

J

s

= −L

11

∇T +L

12

_

E −∇

_

µ

e

z

e

__

, (3.13)

i = −L

21

∇T +L

22

_

E −∇

_

µ

e

z

e

__

, (3.14)

with L

12

= L

21

in virtue of Onsager’s reciprocal relations. For practical

reasons, it is convenient to resolve (3.13) and (3.14) with respect to J

s

and

E −∇(µ

e

/z

e

) and to introduce the following phenomenological coeﬃcients:

λ = T(L

11

−L

12

L

21

/L

22

) (heat conductivity),

π = T(L

12

/L

22

) (Peltier coeﬃcient),

ε = L

21

/L

22

(Seebeck coeﬃcient),

r = 1/L

22

(electrical resistivity).

3.2 Thermoelectric Eﬀects 73

With this notation, (3.13) and (3.14) take the more familiar forms

q = −λ∇T +

_

π +

µ

e

z

e

_

i , (3.15)

E −∇

_

µ

e

z

e

_

i = ε∇T +ri , (3.16)

with, as a consequence of the symmetry property L

12

= L

21

,

π = εT. (3.17)

This result is well known in thermoelectricity as the second Kelvin rela-

tion and has been conﬁrmed experimentally and theoretically in transport

theories.

In most applications, it is permitted to omit the contribution of the terms

in µ

e

/z

e

in relations (3.15) and (3.16), which receive then a simple interpre-

tation. It follows from (3.15) that, even in absence of a temperature gradient,

a heat ﬂux may be generated due to the presence of an electrical current: this

is called the Peltier’s eﬀect. The coeﬃcient π = ±([q[/[i [)

∆T=0

is a measure

of the quantity of heat absorbed or rejected at the junction of two conductors

of diﬀerent materials kept at uniform temperature and crossed by an electric

current of unit density (see Box 3.2).

Expression (3.16) exhibits the property that, even in absence of an electric

current, an electrical potential E = −∇φ can be created due solely to a

temperature diﬀerence, this is known as Seebeck’s eﬀect: the coeﬃcient ε =

±(∆φ/∆T)

i =0

is the Seebeck or thermoelectric power coeﬃcient, it measures

the electrical potential produced by a unit temperature diﬀerence, in absence

of electric current. The Seebeck’s eﬀect ﬁnds a natural application in the

construction of thermocouples (see Box 3.1).

Box 3.1 The Seebeck’s Eﬀect and Thermoelectric Power

The most relevant application of the Seebeck’s eﬀect is the thermocou-

ple, which is frequently used for temperature measurements. The latter is

constituted by two diﬀerent materials (usually metals or alloys) forming

a circuit with some points at diﬀerent temperatures. Consider the situa-

tion presented in Fig. 3.1, where the two junctions between the materials

A and B are at diﬀerent temperatures T

1

and T

2

, respectively. A voltmeter

is inserted inside the system, at a position where temperature is T

0

. In

absence of electric current, the voltmeter indicates a value ∆Φ for the elec-

tric potential diﬀerence, which is related to ∆T ≡ T

2

− T

1

, by means of

∆Φ =

_

T

2

T

1

(ε

A

−ε

B

) dT. Assuming that ε

A

− ε

B

is known and constant,

one has ∆Φ = (ε

A

− ε

B

)∆T, and the measurement of ∆Φ will allow us

to determine the value of ∆T. Usually, one of the two temperatures is a

reference temperature (for instance, the melting point of ice) and the other

one is the temperature to be measured.

74 3 Coupled Transport Phenomena

Fig. 3.1 Thermocouple

The thermoelectric power of the thermocouple is deﬁned as the dif-

ference between the absolute thermoelectric powers of the two materials,

namely ε

AB

= ε

B

− ε

A

. Intuitively, the thermoelectric power reﬂects the

tendency of carriers to diﬀuse from the hot end (characterized by a higher

thermal energy whence higher average kinetic energy of the carriers) to the

cold end, carrying with them their electric charge. The sign of the See-

beck coeﬃcient ε is taken as positive when the carriers are positive charges

(as for instance holes in p-doped semiconductors) and negative otherwise

(as in most metals or n-doped semiconductors). This rule is not respected

in some metals, like copper or silver, which are characterized by a positive

thermoelectric power. The usual values of ε are of the order of 10 µVK

−1

for metals and 100 µVK

−1

for semiconductors. Typical examples of thermo-

couples are those formed by the junctions of copper/constantan (a copper–

nickel alloy), nickel–chromium alloy/constantan, or platinum–rhodium

alloy/platinum.

Another interesting application of the Seebeck’s eﬀect is thermoelectric

generation, namely electrical power generated from a temperature diﬀer-

ence; the discussion of this subject will be postponed later.

Box 3.2 The Peltier’s Eﬀect and Thermoelectric Cooling Systems

The Peltier’s eﬀect describes the absorption or emission of heat accompa-

nying the ﬂow of an electric current across the junctions of two materials

A and B, under isothermal conditions (see Fig. 3.2). Assume, for instance,

that the circulation of the current from A to B produces a cooling of the

junction between them. The eﬀect is reversed by reversing the direction of

the electric current.

To interpret this phenomenon from a microscopic perspective is easy.

If in material A the electrons are moving at an energy level lower than in

material B, the transition of one electron from A to B will require the supply

of some energy from the junction and therefore the passage of current is

accompanied by cooling the right junction. The reverse is true by transiting

from B to A: some quantity of energy is released at the left junction which

3.2 Thermoelectric Eﬀects 75

Fig. 3.2 The Peltier’s eﬀect

is heated. An example is that of electrons ﬂowing from a low energy level in

p-type semiconductors to a higher-energy level in n-type semiconductors.

Materials frequently used are bismuth telluride (Bi

2

Te

3

) heavily doped and,

more recently, the layered oxide Na

x

Co

2

O

4

.

A practical application of the Peltier’s eﬀect is the thermoelectric re-

frigerator. If the junction A → B is cooled, when electric current ﬂows

in a given direction, it will be put inside the container to be refrigerated,

whereas the junction B → A, delivering heat, is left outside. The eﬀect of

the passage of a current is thus to extract heat from the container and to

deliver it to the external world. When current ﬂows in the opposite direc-

tion, the reverse process occurs, namely extracting heat from the external

world and heating the container. The analogy with usual refrigerators or

heat pumps based on Carnot cycle is the following: the junction absorbing

heat is analogous to the expansion of refrigerant in contact with the cold

reservoir whereas the junction delivering heat is the analogous of the com-

pression of the refrigerant in contact with the hot reservoir. Thermoelectric

coolers have the advantage of not having moving parts, since there is neither

compression nor expansion of the refrigerant, nor phase change. They are

reliable, do not need maintenance and can be given a very compact form.

Nevertheless, they present the disadvantage to consume more energy and

are therefore less eﬃcient than usual coolers.

A third thermoelectric coupling of interest is Thomson’s heat. This

eﬀect occurs from the simultaneous presence of a temperature gradient and

an electric current, and is the result of the coupling of both the Peltier’s

and Seebeck’s eﬀects. Substituting (3.15) and (3.16) in the energy balance

equation (3.10) and taking into account of the conservation of total charge

(∇ i = 0) and Kelvin’s second relation, it is found that

ρ

du

dt

= ∇ (λ∇T) +ri

2

−i (∇π)

T

+

_

π

T

−

∂π

∂T

_

i ∇T, (3.18)

76 3 Coupled Transport Phenomena

where subscript T means that the temperature is ﬁxed. If the temperature

remains uniform, only the Joule eﬀect ri

2

and the Peltier’s eﬀect i (∇π)

T

contribute to the dissipated heat. Now, in presence of a temperature gradi-

ent, two supplementary terms occur in (3.18): heat conduction plus a term

resulting from the coupling i ∇T of the electric current and the tempera-

ture gradient and called the Thomson’s heat. The latter may be positive or

negative in contrast with Joule heating, which is invariably positive.

The coeﬃcient

σ

Th

≡

π

T

−

∂π

∂T

(3.19)

is the so-called Thomson coeﬃcient, which, in virtue of π = εT can still be

cast in the form

σ

Th

= −T

∂ε

∂T

. (3.20)

This result is referred to as the ﬁrst Kelvin relation and exhibits the strong

connection between the various thermoelectric eﬀects.

3.2.2 Eﬃciency of Thermoelectric Generators

We have shortly presented in Boxes 3.1 and 3.2 some practical applications

of thermoelectric materials. Motivated by the developments of research on

new materials (Mahan et al. 1997), let us calculate the eﬃciency of thermo-

electric conversion. Consider a single one-dimensional thermoelectric element

of length l under steady conditions. The hot side is at a temperature T

h

(as-

sumed to be the upper side, at y = l), and the cold side (the lower side, at

y = 0) is at temperature T

c

. An electric current i and a quantity of heat

˙

Q enter uniformly into the hot side of the element. The eﬃciency η of the

generator is deﬁned as the ratio of electric power output and heat supplied

per unit time.

The electric power output is the product of the electric current i and

the electric ﬁeld E, the latter being given by (3.16) with the term ∇(µ

e

/z

e

)

neglected:

P

el

= iE = i

_

l

0

ε(T)(dT/dy)dy −i

2

_

l

0

r(T) dy. (3.21)

The ﬁrst term is the Seebeck one and the second one describes Joule’s dissipa-

tion. The sign minus in the second term of the right-hand side arises because

in the present example ∇T and i have opposite directions. The total heat ﬂux

˙

Q supplied per unit time at the hot side is the sum of the purely conductive

contribution given by Fourier’s law and the Peltier term πi from (3.15) when

the term µ

e

/z

e

is again neglected. Making use of the second Kelvin relation

(3.17), one may write

3.2 Thermoelectric Eﬀects 77

˙

Q = λ

h

(T

h

−T

c

)l

−1

+iT

h

ε

h

, (3.22)

where subscript “h” means that the mentioned quantities are evaluated at

T = T

h

, in addition, both

˙

Q and i are supposed to be directed downwards.

The eﬃciency of the thermoelectric generator is given by

η ≡

P

el

˙

Q

≈

iε(T

h

−T

c

) −i

2

rl

λ(T

h

−T

c

)l

−1

+iT

h

ε

. (3.23)

If the diﬀerence of temperatures is not very important, in such a way that

the generator may be considered as almost homogeneous, the quantities λ, ε,

and r may be taken as constants, as it has been done in (3.23).

It is usual to express the eﬃciency η as the product of Carnot’s eﬃciency

η

Carnot

≡ (T

h

−T

c

)/T

h

and a so-called reduced eﬃciency η

r

, i.e. η = η

Carnot

η

r

.

We therefore write (3.23) as

η =

T

h

−T

c

T

h

iε −

ri

2

l

T

h

−T

c

λ

T

h

−T

c

lT

h

+iε

. (3.24)

The reduced eﬃciency may still be expressed in terms of the ratio of ﬂuxes

x = i/[λ(T

h

−T

c

)l

−1

] as

η

r

(x) =

εx −λrx

2

T

−1

h

+εx

=

εx[1 −(xε/Z)]

T

−1

h

+εx

, (3.25)

where Z is the so-called thermoelectric ﬁgure of merit

Z =

ε

2

rλ

, (3.26)

which has the dimension of the reciprocal of temperature, and depends only

on the transport coeﬃcients, generally function of the temperature. Expres-

sion (3.26) of the ﬁgure of merit reﬂects the property that eﬃciency is

enhanced by high values of the Seebeck coeﬃcient ε, low values of the

electric resistivity r, and heat conductivity λ. High values of ε contribute

to a strong coupling between heat and electric current and low values

of r and λ minimize the losses due to Joule’s dissipation and to heat

conduction.

In the numerator of (3.25) are competing the thermoelectric eﬀect with

Joule dissipation. Since the ﬁrst one is linear in the current, or x, and the

second one is non-linear in the same quantity, there exists necessarily an

optimal value of the reduced eﬃciency. The optimum ratio of the ﬂuxes x

opt

corresponding to the maximum value of η

r

is obtained from the condition

dη

r

(x)/dx = 0; the result is

x

opt

=

(1 +ZT)

1/2

−1

εT

. (3.27)

78 3 Coupled Transport Phenomena

The maximum eﬃciency is then

η

max

=

T

h

−T

c

T

h

2 +ZT −2

√

1 +ZT

ZT

. (3.28)

If the ratio x is signiﬁcantly diﬀerent from the optimum value (3.27), the

material is not eﬃciently converting heat energy into electric energy. Once

x is optimized, the optimal current may be found, for a given value of the

temperature gradient. Note that for ZT ¸ 1, one ﬁnds back Carnot’s but

materials currently used in thermoelectric devices have relatively low values

of ZT, between 0.4 and 1.5, so that in reality, the maximum Carnot value is

far from being attained.

In general, the temperature diﬀerence between the two sides of the genera-

tor will be so high that the assumption of homogeneity used in (3.25)–(3.28)

is not tenable. In this case, one may consider the thermoelectric device as

formed by a series of small quasi-homogeneous elements, at diﬀerent average

temperatures. For instance, for two elements thermally in series, the com-

bined eﬃciency is (see Problem 3.9),

η =

P

el,1

+P

el,2

˙

Q

1

= 1 −(1 −η

1

)(1 −η

2

). (3.29a)

In the continuum limit, in which the generator is constituted of many layers

at diﬀerent temperatures, (3.29a) must be replaced by (see Problem 3.9),

η = 1 −exp

_

−

_

T

h

T

c

η

r

(x, T)

T

dT

_

. (3.29b)

Using this expression, one ﬁnds for the total eﬃciency of the device

η = 1 −

ε

c

T

c

+x

−1

c

ε

h

T

h

+x

−1

h

. (3.30)

In this conﬁguration, it may happen that the optimum current determined

by x

opt

in one segment (for instance, the hot side) is signiﬁcantly diﬀerent

from the optimum value x

opt

in another segment (for instance, the cold side);

in this case, there will be no suitable current for which both parts of the gen-

erator are operating with optimal eﬃciency. This is a challenge in materials

sciences, as x

opt

is temperature dependent through the transport coeﬃcients

of the material (namely Z and ε) and it would be highly desirable to ﬁnd ma-

terials with suitable temperature dependence of these coeﬃcients to optimize

the generation.

In terms of the coeﬃcient Z and assuming constant transport coeﬃcients

ε, λ, and r, the maximum eﬃciency of the power generation is given by (see

Problem 3.7)

η

max

=

T

h

−T

c

T

h

(1 +ZT

av

)

1/2

−1

(1 +ZT

av

)

1/2

+ (T

c

/T

h

)

, (3.31)

3.3 Thermodiﬀusion: Coupling of Heat and Mass Transport 79

where T

av

is the average temperature T

av

=

1

2

(T

h

+T

c

). When T

c

≈ T

h

, the

ratio T

c

/T

h

in the denominator of (3.31) is close to 1 and the eﬃciency (3.28)

is recovered.

Analogously, the coeﬃcient of performance for refrigeration (i.e. the ratio

between the heat extracted per unit time and the electric power consumed

by the corresponding engine) is

η =

˙

Q

P

el

=

T

c

T

h

−T

c

(1 +ZT

av

)

1/2

−(T

h

/T

c

)

1 + (1 +ZT

av

)

1/2

. (3.32)

In many practical situations, two parallel generators are used, one of

n-type semiconductors (current carried by electrons, with ε < 0) and an-

other with p-type semiconductors (current brought by holes, with ε > 0).

The global eﬃciency may be derived from (3.23) or (3.31) by using an aver-

age for both generators, namely

η

np

=

η

p

Q

p

+η

n

Q

n

Q

p

+Q

n

, (3.33)

where Q

p

and Q

n

are the amounts of heat supplied to the hot side of the p

and n elements per unit time.

3.3 Thermodiﬀusion: Coupling of Heat and Mass

Transport

By thermodiﬀusion is meant the coupling between heat and matter transport

in binary or multi-component mixtures. In the case of a binary mixture, a

natural choice of the state variables is ρ

1

, ρ

2

(individual mass densities), v

(barycentric velocity), and u (internal energy), but a more convenient choice

is ρ(= ρ

1

+ ρ

2

), c

1

= (ρ

1

/ρ), v, and u. In the classical theory of irreversible

processes, one is more interested by the behaviour of the barycentric velocity

than by the individual velocities of the components and this is the reason why

only the velocity v, and not the individual velocities v

1

and v

2

, ﬁgures among

the space of variables. In absence of chemical reactions, viscosity, external

body forces and energy sources, the corresponding evolution equations are

given by

dρ

dt

= −ρ∇ v, (3.34)

ρ

dc

1

dt

= −∇ J

1

, (3.35)

ρ

dv

dt

= −∇p, (3.36)

ρ

du

dt

= −∇ q −p∇ v. (3.37)

80 3 Coupled Transport Phenomena

Moreover, we assume that the system is in mechanical equilibrium with zero

barycentric velocity and zero acceleration, which is the case for mixtures

conﬁned in closed vessels. It follows then from (3.34) and (3.36) that the

total mass density and the pressure remain uniform throughout the system

and that the last term in (3.37) vanishes. Substituting the balance equations

of mass fraction (3.35) and internal energy (3.37) in the Gibbs’ equation

ds

dt

= T

−1

du

dt

−T

−1

(¯ µ

1

− ¯ µ

2

)

dc

1

dt

, (3.38)

where use has been made of dc

2

= −dc

1

, results in the following balance of

entropy

ρ

ds

dt

= −∇¦T

−1

[q −(¯ µ

1

−¯ µ

2

)J

1

]¦+q ∇T

−1

−J

1

∇[T

−1

(¯ µ

1

−¯ µ

2

)], (3.39)

with the entropy ﬂux given by

J

s

= T

−1

[q −(¯ µ

1

− ¯ µ

2

)J

1

], (3.40)

and the rate of entropy production by

σ

s

= q ∇T

−1

−J

1

∇[T

−1

(¯ µ

1

− ¯ µ

2

)]. (3.41)

With the help of the Gibbs–Duhem’s relation

c

1

(∇¯ µ

1

)

T,p

+c

2

(∇¯ µ

2

)

T,p

= 0, (3.42)

where subscripts T and p indicate that diﬀerentiation is taken at constant T

and p, and the classical result of equilibrium thermodynamics

T∇(T

−1

¯ µ

k

) = −h

k

T

−1

(∇T) + (∇¯ µ

k

)

T

, (3.43)

with h

k

the partial speciﬁc enthalpy of substance k(k = 1, 2), we are able to

eliminate the chemical potential from (3.41) of σ

s

, which ﬁnally reads as

σ

s

= −q

∇T

T

2

−

µ

11

Tc

2

J

1

∇c

1

. (3.44)

The new heat ﬂux q

is deﬁned by q

= q −(h

1

−h

2

)J

1

, and it is equal to the

diﬀerence between the total ﬂux of heat and the transfer of heat due to diﬀu-

sion while the quantity µ

11

stands for µ

11

= (∂¯ µ

1

/∂c

1

)

T,p

. The derivation of

(3.44) exhibits clearly the property that the entropy production is a bilinear

expression in the thermodynamic ﬂuxes q

and J

1

and forces taking the form

of gradients of intensive variables, easily accessible to direct measurements.

It is obvious that when the mixture reaches thermodynamic equilibrium, the

heat and mass ﬂows as well as the temperature and mass fraction gradients

3.3 Thermodiﬀusion: Coupling of Heat and Mass Transport 81

vanish. Expression (3.44) of σ

s

suggests writing the following phenomenolog-

ical relations between ﬂuxes and forces:

q

= −L

qq

∇T

T

2

−L

q1

µ

11

Tc

2

∇c

1

, (3.45)

J

1

= −L

1q

∇T

T

2

−L

11

µ

11

Tc

2

∇c

1

, (3.46)

with the Onsager’s reciprocal relation L

q1

= L

1q

and the following inequali-

ties resulting from the positiveness of entropy production:

L

qq

> 0, L

11

> 0, L

qq

L

11

−L

q1

L

1q

> 0. (3.47)

After introducing the following identiﬁcations:

L

qq

T

2

= λ (heat conductivity),

L

11

µ

11

ρc

2

T

= D (diﬀusion coeﬃcient),

L

q1

ρc

1

c

2

T

2

= D

F

(Dufour coeﬃcient),

L

1q

ρc

1

c

2

T

2

= D

T

(thermal diﬀusion coeﬃcient),

the phenomenological laws take the form

q

= −λ∇T −ρTµ

11

c

1

D

F

∇c

1

, (3.48)

J

1

= −ρc

1

c

2

D

T

∇T −ρD∇c

1

. (3.49)

Inequalities (3.47) imply in particular that λ > 0 and D > 0 while from

Onsager’s relation is inferred that

D

F

= D

T

. (3.50)

This last result is a conﬁrmation of an earlier result established by Stefan

at the end of the nineteenth century. Starting from the law of conservation

of momentum, Stefan was indeed able to demonstrate the above equality at

least in the case of binary mixtures. By setting the gradient of the mass

fraction equal to zero, (3.48) is identical to Fourier’s equation so that λ can

be identiﬁed with the heat conductivity coeﬃcient. Similarly, relation (3.49)

reduces to Fick’s law of diﬀusion when temperature is uniform and there-

fore D represents the diﬀusion coeﬃcient; in general, the phenomenological

coeﬃcients in (3.48) and (3.49) are not constant.

In multi-component mixtures, the mass ﬂux of substance i is a linear

function of not only ∇c

i

but also of all the other mass fractions gradients

∇c

j

(j ,= i). Such “diﬀusion drag” forces have been invoked to interpret some

biophysical eﬀects and play a role in the processes of separation of isotopes

(see Box 3.3).

82 3 Coupled Transport Phenomena

Box 3.3 The Soret’s Eﬀect and Isotope Separation

Thermal diﬀusion is exploited to separate materials of diﬀerent molecular

mass. If a ﬂuid system is composed of two kinds of molecules of diﬀerent

molecular weight, and if it is submitted to a temperature gradient, the

lighter molecules will accumulate near the hot wall and the heavier ones near

the cold wall. This property was used in the 1940s for the separation of

235

U

and

238

U isotopes in solutions of uranium hexaﬂuoride, in the Manhattan

project, leading to the ﬁrst atomic bomb. Usually this process is carried

out in tall and narrow vertical columns, where convection eﬀects reinforce

the separation induced by thermal diﬀusion: the light molecules near the

hot walls have an ascending motion, whereas the heavy molecules near

the cold wall sunk towards the lowest regions. This process accumulates

the lightest isotope in the highest regions, from where it may be extracted.

This process is simple but its consumption of energy is high, and therefore

it has been substituted by other methods. However, it is still being used in

heavy water enrichment, or in other processes of separation of light atoms

with the purpose of, for instance, to generate carbide layers on steel, alloys

or cements, thus hardening the surface and making it more resistant to wear

and corrosion. Soret’s eﬀect plays also a role in the structure of ﬂames and in

polymer characterization. More recently, high temperature gradients have

been produced by means of laser beams rather than by heating uniformly

the walls of the container.

The Dufour’s eﬀect, the reciprocal of thermal diﬀusion, has not so many

industrial applications. It plays nevertheless a non-negligible role in some

natural processes as heat transport in the high atmosphere and in the soil

under isothermal conditions but under a gradient of moisture.

The coeﬃcient D

T

in (3.49) is typical of thermal diﬀusion, i.e. the ﬂow of

matter caused by a temperature diﬀerence; such an eﬀect is referred to as the

Soret’s eﬀect in liquids with the quotient D

T

/D called the Soret coeﬃcient.

The reciprocal eﬀect, i.e. the ﬂow of heat caused by a gradient of concen-

tration as evidenced by (3.48) is the Dufour’s eﬀect. It should be observed

that the cross-coeﬃcients D

T

and D

F

are much smaller than the direct co-

eﬃcients like the heat conductivity λ and the diﬀusion coeﬃcient D. The

latter turns out to be of the order of 10

−8

m

2

s

−1

in liquids and 10

−5

m

2

s

−1

in gases while the coeﬃcient of thermal diﬀusion D

T

varies between 10

−12

and 10

−14

m

2

s

−1

K

−1

in liquids and from 10

−8

to 10

−12

m

2

s

−1

K

−1

in gases.

The Soret’s eﬀect is mainly observed in oceanography while the Dufour’s

eﬀect, which is negligible in liquids, has been detected in the high atmosphere.

The smallness of the coupling eﬀects is the reason why they are hard to be

observed and measured with accurateness.

Deﬁning a stationary state by the absence of matter ﬂow (J

1

= 0), it turns

out from (3.49) that

(∆T)

st

= −

D

D

T

c

1

c

2

(∆c

1

)

st

, (3.51)

3.4 Diﬀusion Through a Membrane 83

which indicates that a diﬀerence of concentration is able to generate a tem-

perature diﬀerence, called the osmotic temperature. This is typically an irre-

versible eﬀect because the corresponding entropy production is non-zero as

directly seen from (3.41). This eﬀect should not be confused with the os-

motic pressure, which expresses that a diﬀerence of concentration between

two reservoirs kept at the same uniform temperature but separated by a

semi-permeable membrane gives raise to a pressure drop, called the osmotic

pressure. The latter is a pure equilibrium eﬀect resulting from the property

that, at equilibrium, the chemical potential ¯ µ(T, p, c

1

) takes the same value

in both reservoirs so that (∆¯ µ)

T

= ∆p/ρ

1

+µ

11

∆c

1

= 0, and

∆p = −ρ

1

µ

11

∆c

1

, (3.52)

with ρ

1

the speciﬁc mass of the species crossing the membrane; it is directly

checked that in the present situation, the entropy production (3.41) is indeed

equal to zero.

The phenomena studied in this section are readily generalized to multi-

component electrically charged systems, like electrolytes.

3.4 Diﬀusion Through a Membrane

The importance of transport of matter through membranes in the life of cells

and tissues justify that we spend some time to discuss the problem. In biolog-

ical membranes, one distinguishes generally between two modes of transport:

purely passive transport due to a pressure gradient or a mass concentration

gradient and active transport involving ionic species, electrical currents, and

chemical reactions. Here we focus on some aspects of passive transport. Our

objective is to present a simpliﬁed analysis by using a minimum number of

notions and parameters; in that respect, thermal eﬀects will be ignored but

even so, the subject keeps an undeniable utility.

We consider the simple arrangement formed by two compartments I and

II separated by a homogeneous membrane of thickness ∆l, say of the order

of 100 µm. Each compartment is ﬁlled with a binary solution consisting of a

solvent 1 and a solute 2 (see Fig. 3.3).

The membrane is assumed to divide the system in two discontinuous sub-

systems that are considered as homogeneous.

3.4.1 Entropy Production

In absence of thermal gradients, it is inferred from (3.41) that the rate of

dissipation, measured per unit volume of the membrane, will take the form

84 3 Coupled Transport Phenomena

Fig. 3.3 System under study consisting in a membrane separating a binary solution

Tσ

s

= −J

1

∇¯ µ

1

−J

2

∇¯ µ

2

. (3.53)

After integration over the thickness ∆l, the rate of dissipation per unit sur-

face, denoted as Φ, can be written as

Φ = −J

1

∆¯ µ

1

−J

2

∆¯ µ

2

, (3.54)

where ∆¯ µ

i

designates the diﬀerence of chemical potential of species i across

the membrane, J

1

and J

2

are the ﬂows of solvent and solute, respectively.

Instead of working with J

1

and J

2

, it is more convenient to introduce the

total volume ﬂow J

V

across the membrane and the relative velocity J

D

of the

solute with respect to the solvent, deﬁned, respectively, by

J

V

= ¯ v

1

J

1

+ ¯ v

2

J

2

, (3.55)

J

D

= v

2

−v

1

, (3.56)

the quantities ¯ v

1

and ¯ v

2

stand for the partial speciﬁc volumes of the solvent

and the solute, v

1

and v

2

are their respective velocities given by v

1

= ¯ v

1

J

1

and

v

2

= ¯ v

2

J

2

. With the above choice of variables, (3.54) reads as (Katchalsky

and Curran 1965; Caplan and Essig 1983)

Φ = −J

V

∆p −J

D

∆π, (3.57)

where ∆p = p

I

−p

II

and ∆π = c

2

(∆¯ µ

2

)

p

is the osmotic pressure; the quantity

(∆¯ µ

2

)

p

is that part of the chemical potential depending only on the concen-

tration and deﬁned from ∆¯ µ

2

= V

2

∆p + (∆¯ µ

2

)

p

, c

2

is the number of moles

of the solute per unit volume. For ideal solutions, one has (∆¯ µ

2

)

p

= RT∆c

2

.

3.4 Diﬀusion Through a Membrane 85

3.4.2 Phenomenological Relations

By assuming linear relations between thermodynamic ﬂuxes and forces, one

has

J

V

= L

VV

∆p +L

VD

∆π, (3.58)

J

D

= L

DV

∆p +L

DD

∆π. (3.59)

The advantage of (3.58) and (3.59) is that they are given in terms of parame-

ters that are directly accessible to experiments. The corresponding Onsager’s

relation is L

VD

= L

DV

, whose main merit is to reduce the number of para-

meters from four to three.

To better apprehend the physical meaning of the phenomenological coeﬃ-

cients L

VV

, L

DV

, L

DD

, and L

VD

, let us examine some particular experimental

situations. First consider the case wherein the concentration of the solute is

the same on both sides of the membrane such that ∆π = 0. If a pressure

diﬀerence ∆p is applied, one will observe according to (3.58) a volume ﬂow

proportional to ∆p; the proportionality coeﬃcient L

VV

is called the mechan-

ical ﬁltration coeﬃcient of the membrane: it is deﬁned as the volume ﬂow

produced by a unit pressure diﬀerence between the two faces of the mem-

brane. A further look on relation (3.59) indicates that even in absence of a

concentration diﬀerence (∆π = 0), there will be a diﬀusion ﬂow J

D

= L

DV

∆p

caused by the pressure diﬀerence ∆p. This phenomenon is known in colloid

chemistry under the name of ultraﬁltration and the coeﬃcient L

DV

is the

ultraﬁltration coeﬃcient. An alternative possibility is to impose ∆p = 0 but

diﬀerent solute concentrations in compartments I and II. In virtue of (3.59),

the osmotic diﬀerence ∆π will produce a ﬂow of diﬀusion J

D

= L

DD

∆π and

L

DD

is identiﬁed as the permeability coeﬃcient: it is the diﬀusional mobility

induced by a unit osmotic pressure ∆π = 1. Another eﬀect related to (3.58)

is the occurrence of a volume ﬂow J

V

= L

VD

∆π caused by a diﬀerence of os-

motic pressure at uniform hydrostatic pressure: the coupling coeﬃcient L

VD

is referred to as the coeﬃcient of osmotic ﬂow.

The above discussion has clearly shown the importance of the coupling

coeﬃcient L

DV

= L

VD

; by ignoring it one should miss signiﬁcant features

about motions across membranes. The importance of this coeﬃcient is still

displayed by the osmotic pressure experiment illustrated by Fig. 3.4.

The two phases I (solvent +solute) and II (solvent alone) are separated by

a semi-permeable membrane only permeable to the solvent. The height of the

solution in the capillary tube gives a measure of the ﬁnal pressure diﬀerence

obtained when the volume ﬂow J

V

vanishes, indeed from (3.58) one obtains

(∆p)

J

V

=0

= −

L

VD

L

VV

∆π. (3.60)

This result indicates that, contrary to what is sometimes claimed, ∆p is

not a measure of the osmotic pressure, this is only true if L

VD

= −L

VV

.

86 3 Coupled Transport Phenomena

Fig. 3.4 Osmotic pressure experiment

This condition is met by so-called ideal semi-permeable membranes whose

property is to forbid the transport of solute whatever the values of ∆p and

∆π. For membranes, which are permeable to the solute, it is experimentally

found that

r ≡ −

L

VD

L

VV

< 1. (3.61)

This ratio that is called the reﬂection coeﬃcient tends to zero for selec-

tive membranes, like porous gas ﬁlters, and has been proposed to act as a

measure of the selectivity of the membrane. For r = 1 (ideal membranes), all

the solute is reﬂected by the membrane, for r < 1, some quantity of solute

crosses the membrane, while for r = 0 the membrane is completely permeable

to the solute. To clarify the notion of membrane selectivity, let us go back to

the situation described by ∆π = 0. In virtue of (3.58) and (3.59), one has

r = −

L

VD

L

VV

= −

_

J

D

J

V

_

∆π=0

, (3.62)

or, in terms of the velocities introduced in relation (3.56),

_

2v

2

v

1

+v

2

_

∆π=0

= 1 −r. (3.63)

For an ideal semi-permeable membrane (r = 1), one has v

2

= 0 and the

solute will not cross the membrane; for r = 0 (v

1

= v

2

) the membrane is

not selective and allows the passage of both the solute and the solvent; for

negative values of r(v

2

> v

1

), the velocity of the solute is greater than that

of the solvent and this is known as negative anomalous osmosis, which is

a characteristic of the transport of electrolytes across charged membranes.

3.5 Problems 87

Table 3.1 Phenomenological coeﬃcients for two biological membranes

ω L

VV

Membrane Solute Solvent (10

−15

mol dyn

−1

s

−1

) r (10

−11

cm

3

dyn

−1

s

−1

)

Human Methanol Water 122 – –

blood cell

Urea Water 17 0.62 0.92

Toad skin Acetamide Water 4 ×10

−3

0.89 0.4

Thiourea Water 5.7 ×10

−4

0.98 1.1

A ﬁnal parameter of interest, both in synthetic and biological membranes, is

the solute permeability coeﬃcient

ω =

c

2

L

VV

(L

VV

L

DD

−L

2

VD

). (3.64)

For ideal semi-permeable membranes for which L

VD

= −L

VV

= −L

DD

,

one has ω = 0, and for non-selective membranes (r = 0), it is found that

ω = c

2

L

DD

.

The interest of irreversible thermodynamics is to show clearly that three

parameters are suﬃcient to describe transport of matter across membrane

and to provide the relationships between the various coeﬃcients character-

izing a semi-permeable membrane. In Table 3.1 are reported some values

of these coeﬃcients for two diﬀerent biological membranes (Katchalsky and

Curran 1965).

3.5 Problems

3.1. Entropy ﬂux and entropy production. Determine (3.11) and (3.12) of the

entropy ﬂux and the entropy production in the problem of thermoelectricity.

3.2. Onsager’s reciprocal relations. In presence of a magnetic ﬁeld, Onsager’s

relations can be written as L(H) = L

T

(−H). Decomposing L in a symmetric

and an antisymmetric part L = L

s

+ L

a

, show that L

s

(H) = L

s

(−H) and

L

a

(H) = −L

a

(−H).

3.3. Thermoelectric eﬀects. The Peltier coeﬃcient of a couple Cu–Ni is

π

Cu–Ni

≈ −5.08 mV at 273 K, π

Cu–Ni

≈ −6.05 mV at 295 K, and π

Cu–Ni

≈

−9.10 mV at 373 K. Evaluate (a) the heat exchanged in the junction by

Peltier’s eﬀect when an electric current of 10

−2

A ﬂows from Cu to Ni at

295 K; (b) idem when the current ﬂows from Ni to Cu. (c) The two junctions

of a thermocouple made of Cu and Ni are kept at 305 and 285 K, respec-

tively; by using the Thomson relation, determine the Seebeck coeﬃcient and

estimate the electromotive force developed by the thermocouple.

88 3 Coupled Transport Phenomena

3.4. Thermocouple. The electromotive force ∆ξ of a Cu–Fe thermocouple is

given by ∆ξ(µV ) ≈ −13.4∆T − 0.01375(∆T)

2

, where ∆T = T

R

− T, the

reference temperature being T

R

= 273 K. (a) Evaluate the Peltier heat ex-

changed in a Cu–Fe junction at 298 K when an electric current of 10

−3

A ﬂows

from Cu to Fe. (b) By using the Thomson relation, evaluate the diﬀerence of

Thomson coeﬃcients of these two metals at 298 K.

3.5. Evaluation of the maximum eﬃciency of a thermoelectric element.

Assume a ﬁnite temperature diﬀerence and constant transport coeﬃcients.

(a) Starting from the deﬁnition of the ratio of ﬂuxes x ≡ i/λ∇T, and from

the steady state form of (3.18), prove that

dx

−1

dT

= T

dε

dT

−xλr.

(b) Taking into account this relation, and for ε independent of T, show that

the values of x at the cold and hot boundaries of the systems, x

c

and x

h

,

respectively, may be written in terms of x

opt,av

(Snyder and Ursell 2003)

x

−1

h

= x

−1

opt,av

−

1

2

∆T(λr)x

opt,av

,

x

−1

c

= x

−1

opt,av

+

1

2

∆T(λr)x

opt,av

,

where x

opt,av

is the value for x

opt

found in (3.27) for the average temperature

T

av

=

1

2

(T

h

+T

c

). (c) Introducing these values in (3.30), show ﬁnally (3.31),

i.e.

η =

T

h

−T

c

T

h

(1 +ZT

av

)

1/2

−1

(1 +ZT

av

)

1/2

+ (T

c

/T

h

)

.

3.6. The ﬁgure of merit of a thermoelectric material. The ﬁgure of merit

for thermoelectric materials is deﬁned by Z ≡ ε

2

/λr, where ε is the Seebeck

coeﬃcient, r is the electrical resistivity, and λ is the thermal conductivity. (a)

Check that the dimension of this combination is the inverse of temperature.

(b) The ﬁgure of merit of a junction of two thermoelectric materials in a

thermocouple is deﬁned as

Z

tc

≡

ε

2

AB

[(λ/σ)

1/2

A

+ (λ/σ)

1/2

B

]

2

.

Show that the eﬃciency of the thermocouple, deﬁned as the ratio of the

electric power delivered and the rate of heat supplied to the hot junction is

η

tc

=

T

1

−T

2

3T

1

+T

2

2

+

4

Z

tc

.

3.7. Heat engines in series. Consider two Carnot heat engines in series: the

ﬁrst one works between heat reservoirs at T

1

and T

2

and the second one

between reservoirs at T

2

and T

3

, assume in addition that the whole amount

3.5 Problems 89

of heat Q

2

delivered by the ﬁrst engine to the reservoir T

2

is transferred to

the second engine. (a) Show that one recovers (3.31), namely

η =

P

el,1

+P

el,2

Q

1

= 1 −(1 −η

1

)(1 −η

2

).

(b) In (3.29b) it has been considered the continuum limit of high number of

Carnot engines working between reservoirs in series at temperatures T

i

+dT

and T

i

, with T

i

the temperature of the cold reservoir corresponding to ith

Carnot engine, T

i

ranging from T

c

to T

h

. Show that the previous expression

leads to (3.29b) by writing

η = 1 −

∞

i=1

(1 −η

i

) ≈ 1 −exp

_

−

_

T

h

T

c

η(T)

T

dT

_

.

3.8. Thermophoresis. Thermophoresis in a quiescent ﬂuid is described by the

phenomenological equation

J = −D[∇c +S

T

c(1 −c)∇T],

where D is the diﬀusion coeﬃcient and S

T

≡ D

T

/D is the Soret coeﬃcient,

D

T

being the thermal diﬀusion coeﬃcient. (a) Show that for concentrations

much less than unity, the concentration of the solute at temperature T is

given by

c = c

0

exp[−S

T

(T −T

0

)],

where c

0

is the concentration at temperature T

0

. For a DNA sample, it was

found that S

T

≈ 0.14 K

−1

around T = 297 K (Braun and Libchaber 2002).

(b) Calculate c/c

0

for two regions with a temperature diﬀerence ∆T = 2.5 K.

3.9. Cross-eﬀects in membranes. Assume that the ﬂows of water J

1

and solute

J

2

across a membrane are, respectively, expressed as

J

1

= L

11

∆p +L

12

∆c

2

,

J

2

= L

21

∆p +L

22

∆c

2

,

where ∆p and ∆c

2

are, respectively, the diﬀerences of pressure and solute

concentration between both sides of the membrane. Is the Onsager’s recipro-

cal relation L

12

= L

21

applicable? Explain.

3.10. Transport of charged ions across membranes. The Nernst equation.

Transport of charged ions, mainly H

+

, Na

+

, K

+

, Ca

2+

, and Cl

−

, across mem-

branes plays a crucial role in many biological processes. Consider a membrane

with both sides at concentrations c

in

, c

out

, and voltages 1

in

, 1

out

. An argu-

ment similar to that leading to (3.58) and (3.59) yields, in the isothermal

case, the following result for the diﬀusion ﬂux J (number of ions which cross

the membrane per unit area and unit time)

90 3 Coupled Transport Phenomena

J = −

˜

D(∇µ +q∇1),

q designating the electrical charge. (a) Show that the voltage diﬀerence for

which both sides of the membrane are at equilibrium (J = 0) with respect

to the transport of a given ionic species i (assumed to behave as an ideal

substance, i.e. µ = µ

(0)

+k

B

T ln c) is given by

(1

in

−1

out

)

eq

=

k

B

T

q

ln

c

out

c

in

.

This equation is known as Nernst equation. (b) Assume that the membrane is

immersed in solutions of NaCl–water with concentration 1 M (where M stands

for mole per litre) and 0.1 M and that only Na

+

may cross the membrane.

Calculate the diﬀerence of voltage between both sides of the membrane at

equilibrium. (c) In typical biological cells at rest, the K

+

concentration are

c

in

= 150 mM, c

out

= 5 mM, and the voltages are V

out

= 0 mV, V

in

= −70 mV.

Evaluate the Nernst potential for K

+

at T = 310 K, and determine whether

K

+

will move towards or outwards the cell.

3.11. Transport of charged ions across membranes: the Goldmann equation.

In the derivation of the Nernst equation, it is assumed that the membrane

is only permeable to one ionic species. In actual situations, several ionic

species may permeate through the membrane. If one considers the transport

of several ionic species as K

+

, Na

+

, and Cl

−

, which are the most usual ones,

the equilibrium voltage diﬀerence is given by the Goldmann equation

(1

in

−1

out

)

eq

=

k

B

T

q

ln

P

K

c

K,out

+P

Na

c

Na,out

+P

Cl

c

Cl,in

P

K

c

K,in

+P

Na

c

Na,in

+P

Cl

c

Cl,out

,

where P

K

, P

Na

, and P

Cl

refer to the relative permeabilities of the correspond-

ing ions across the membrane. Note that when only one species crosses the

membrane (i.e. when the permeabilities vanish for two of the ionic species),

the Goldmann equation reduces to the Nernst relation. (a) Derive the Gold-

mann equation. Hint: Take into account the electroneutrality condition. (b)

It is asked why the Cl concentrations appear in the equation in a diﬀerent

way as the concentrations of the other two ions, namely c

in

instead of c

out

and vice versa. (c) In the axon at rest, i.e. in the long cylindrical terminal

of the neurons along which the output electrical signals may propagate, the

permeabilities are P

K

≈ 25P

Na

≈ 2P

Cl

, whereas at the peak of the action

potential (the electric nervous signal), P

K

≈ 0.05P

Na

≈ 2P

Cl

. The dramatic

increase in the sodium permeability is due to the opening of sodium chan-

nels when the voltage diﬀerence is less than some critical value. Evaluate

the Goldmann potential in both situations, by taking for the concentrations:

c

K,in

= 400 mM, c

K,out

= 20 mM, c

Na,in

= 50 mM, c

Na,out

= 440 mM, and

c

Cl,in

= 50 mM, c

Cl,out

= 550 mM.

Chapter 4

Chemical Reactions and Molecular

Machines

Eﬃciency of Free-Energy Transfer and Biology

Chemical reactions are among the most widespread processes inﬂuencing life

and economy. They are extremely important in biology, geology, environmen-

tal sciences, and industrial developments (energy management, production

of millions of diﬀerent chemical species, search for new materials). Chemical

kinetics is a very rich but complex topic, which cannot be fully grasped by

classical irreversible thermodynamics, whose kinetic description is restricted

to the linear regime, not far from equilibrium.

Despite this limitation, this formalism is able to yield useful results, es-

pecially when dealing with coupled reactions. Indeed, the study of coupled

processes is one of the most interesting features of non-equilibrium thermo-

dynamics. In particular, the various couplings – between several chemical

reactions, between chemical reactions and diﬀusion, and between chemical

reactions and active transport processes in biological cells – receive a simple

and uniﬁed description. Non-equilibrium thermodynamics is able to provide

clarifying insights to some conceptually relevant problems, particularly in

biology, which are rarely investigated in other formalisms.

In that respect, in the forthcoming we will focus on the problem of ef-

ﬁciency of free-energy transfer between coupled reactions, and the descrip-

tion of biological molecular motors. These engines may be modelled by some

particular cycles of chemical reactions, generalizing the triangular chemical

scheme proposed by Onsager in his original derivation of the reciprocal re-

lations. Special attention will also be devoted to the eﬀects arising from the

combination of chemical reactions and diﬀusion. From a biological perspec-

tive, this coupling provides the basis of cell diﬀerentiation in the course of

development of living organisms. The ingredients of this extremely complex

phenomenon, leading to spatial self-organization far from equilibrium, may

be qualitatively understood by considering a simpliﬁed version of the mech-

anism of coupling between autocatalytic chemical reactions and diﬀusion.

Biological applications of non-equilibrium thermodynamics have been the

subject of several books (Katchalsky and Curran 1965; Nicolis and Prigogine

1977; Hill 1977; Caplan and Essig 1983; Westerhoﬀ and van Dam 1987; Jou

91

92 4 Chemical Reactions and Molecular Machines

and Llebot 1990; Nelson 2004). Here, we present a synthesis of the main ideas

and, for pedagogical purpose, we start the analysis with the problem of one

single chemical reaction.

4.1 One Single Chemical Reaction

Equilibrium thermodynamics provides a general framework to analyse the

equilibrium conditions of chemical reactions, and it leads in a natural way

to the concept of equilibrium constant and its modiﬁcations under changes

of temperature and pressure, as recalled in Sect. 1.7. A clear understanding of

these eﬀects is crucial to establish the range of temperature and pressure for

the optimization of industrial processing. However, this optimization cannot

be carried out without taking into account kinetic eﬀects. For instance, low-

ering temperature has generally as a consequence to slow down the reaction

velocity. On the other side, it is known from equilibrium thermodynamics

that, for exothermic reactions, the eﬃciency of conversion of reactants into

products increases when temperature is lowered. Thus, lowering tempera-

ture has two opposite eﬀects: a lowering of the velocity of reactions and an

increase of eﬃciency. Optimization will therefore result from a compromise

between equilibrium factors and kinetic factors. We will return to this topic

in Chap. 5. Here, we emphasize some kinetic aspects of chemical reactions,

from the point of view of non-equilibrium thermodynamics.

To illustrate our approach consider, as in Chap. 1, the reaction of synthesis

of hydrogen chloride

H

2

+ Cl

2

2HCl. (4.1)

Since we will work in terms of local quantities, it is convenient to introduce

the mass fractions c

k

deﬁned as c

k

≡ m

k

/m, with m

k

the mass of component

k and m the total mass. In virtue of the law of deﬁnite proportions (1.70),

the change of c

k

during the time interval dt may be written as

ρ

dc

k

dt

= ν

k

M

k

dξ

dt

, k = 1, 2, . . ., n, (4.2)

with ν

k

the stoichiometric coeﬃcient of species k, M

k

its molar mass, and

dξ/dt the velocity of reaction per unit volume. This equation can be deduced

from (1.70) when the mole numbers are expressed in terms of mass fractions

and the degree of advancement is given per unit volume.

Besides the mass of the n species, the other relevant variable is the speciﬁc

internal energy u, because of the exchange of heat with the outside. The time

evolution equation for the internal energy u is given by the ﬁrst law

ρ

du

dt

= −∇ q, (4.3)

4.1 One Single Chemical Reaction 93

where q is the heat ﬂux vector expressing the amount of heat exchanged

between the system and the outside world acting as a reservoir.

Now, we want to obtain expressions for the velocity of the reaction in

terms of the corresponding thermodynamic force. This will be achieved within

the general framework set out in Chap. 2, demanding that we previously

determined the expression of the entropy production. To determine it, we

start from the Gibbs’ relation

ds

dt

=

1

T

du

dt

−

k

¯ µ

k

T

dc

k

dt

, (4.4a)

where ¯ µ

k

is the chemical potential of the species k measured per unit mass.

Making use of (4.2) and the relation M

k

¯ µ

k

= µ

k

, with µ

k

the chemical

potential per unit mole, one obtains

ρ

ds

dt

=

1

T

ρ

du

dt

−

k

ν

k

µ

k

T

dξ

dt

. (4.4b)

We know from Sect. 1.7 that for ideal systems, i.e. mixture of ideal gases or

dilute solutions, the chemical potential per unit mole µ

k

is given by

µ

k

= µ

(0)

k

(T, p) +RT ln x

k

, (4.5)

where x

k

≡ N

k

/N is the mole fraction while µ

(0)

k

is independent of the

composition.

At this stage, it is relevant to introduce the notion of aﬃnity deﬁned in

(1.76) as

/ ≡ −

k

ν

k

µ

k

. (4.6)

As seen below, the reaction will proceed forwards if / > 0 or backwards

if / < 0: hence the name of aﬃnity. At ﬁxed values of temperature and

pressure, / is only function of the mole fractions of species, and it was shown

in Sect. 1.7 that

/ = RT ln

K(T, p)

x

ν

1

1

x

ν

2

2

, (4.7a)

with K(T, p) the equilibrium constant. Note that this constant may also be

written as K(T, p) = (x

ν

1

1

)

eq

(x

ν

2

2

)

eq

, because at equilibrium / = 0 and

then the term inside the logarithm must be 1. It is useful, for further purposes

in this chapter, to rewrite (4.7a) as

/ = RT ln

(x

ν

1

1

)

eq

(x

ν

2

2

)

eq

x

ν

1

1

x

ν

2

2

. (4.7b)

A third way of expressing the aﬃnity is in terms of the molar concentrations,

i.e. in the number of moles per unit volume, namely ˜ c

k

≡ N

k

/V . In this case,

(4.7b) may be expressed as

94 4 Chemical Reactions and Molecular Machines

/ = RT ln

(˜ c

ν

1

1

)

eq

(˜ c

ν

2

2

)

eq

˜ c

ν

1

1

˜ c

ν

2

2

. (4.7c)

To derive the expression of the rate of entropy production, we introduce

the balance equations of mass (4.2) and energy (4.3) in Gibbs’ relation (4.4a)

and (4.4b). Since in chemical reactions, the temperature is generally assumed

to remain uniform, one obtains

ρ

ds

dt

= −∇

q

T

+

/

T

dξ

dt

. (4.8)

By comparing this expression with the general form of the entropy balance

equation ρds/dt = −∇ J

s

+σ

s

, it follows that the expression of the entropy

ﬂux is

J

s

=

q

T

, (4.9)

while the entropy production is identiﬁed as

σ

s

=

/

T

w > 0, (4.10)

i.e. a bilinear expression in the thermodynamic force //T and the ﬂux w =

dξ/dt, which is the velocity of reaction. Truly, the reaction must be considered

as the net eﬀect of a forward ﬂux w

+

going from reactants to products minus

a backward ﬂux w

−

going from products to reactants, namely w = w

+

−w

−

.

According to the second law, σ

s

is always a positive quantity and zero

in equilibrium. Thus when / is positive (respectively, negative), the velocity

w = w

+

−w

−

is also positive (negative) and the reaction will take place from

left to right (right to left). At equilibrium, the aﬃnity / is zero and the net

ﬂux w will be zero, i.e. the ﬂux w

+

towards the right is equal to the ﬂux w

−

towards the left.

For a single chemical reaction, the phenomenological relation between ﬂux

and force in the linear regime is

w = l

/

T

, (4.11)

where l is a phenomenological coeﬃcient, which in (4.16) is related to the

microscopic rate constants. On the condition that l > 0, this is the simplest

way to guarantee the positiveness of the corresponding entropy production

given by

σ

s

= l

_

/

T

_

2

. (4.12)

However, it should be stressed that the linear ﬂux–force relation (4.11) is

only valid close to equilibrium. This is easily seen by comparing this result

with the classical kinetic expression for the reaction rate which, in the case

of the synthesis of HCl, reads as

w = w

+

−w

−

= k

+

˜ c

H

2

˜ c

Cl

2

−k

−

˜ c

2

HCl

= k

+

˜ c

H

2

˜ c

Cl

2

_

1 −

1

K

˜ c

2

HCl

˜ c

H

2

˜ c

Cl

2

_

, (4.13)

4.1 One Single Chemical Reaction 95

where k

+

and k

−

are, respectively, the rate constants corresponding to the for-

ward and backward reactions, and K = k

+

/k

−

= (˜ c

2

HCl

)

eq

/[(˜ c

H

2

)

eq

(˜ c

Cl

2

)

eq

].

In terms of the aﬃnity, and in virtue of (4.7c), relation (4.13) can be ex-

pressed as

w = k

+

˜ c

H

2

˜ c

Cl

2

[1 −exp(−//RT)] . (4.14)

For reactions near equilibrium, the aﬃnity / is very small and it is justiﬁed

to make a Taylor expansion around / = 0 so that

w = k

+

(˜ c

H

2

)

eq

(˜ c

Cl

2

)

eq

R

/

T

. (4.15)

This result indicates clearly that the linear ﬂux–force law (4.11) is only true

very close to equilibrium and that the classical theory of irreversible processes

does not cope with situations far from equilibrium. By comparison of the

phenomenological relation (4.11) with the kinetic corresponding law (4.15),

we are able to identify the phenomenological coeﬃcient l as

l = k

+

(˜ c

H

2

)

eq

(˜ c

Cl

2

)

eq

R

, (4.16)

which conﬁrms that l is a positive quantity as all the factors of (4.16) are

positive.

The phenomenological relation (4.11) allows us to determine the behaviour

of the degree of advancement ξ in the course of time. Indeed, if / is only

depending on ξ, one can write (4.11) under the form

dξ

dt

=

l

T

/ =

l

T

_

/

eq

+

_

∂/

∂ξ

_

(ξ −ξ

eq

)

_

=

l

T

_

∂/

∂ξ

_

(ξ −ξ

eq

), (4.17)

after use is made of /

eq

= 0. This result suggests deﬁning a relaxation time

τ by

τ ≡ −

T

l(∂//∂ξ)

eq

> 0. (4.18)

In this expression, there appear two diﬀerent factors: one of them is the kinetic

phenomenological coeﬃcient l, whose positiveness is demanded by the second

law; the other one (∂//∂ξ)

T

is related to the thermodynamic stability of the

system and is negative, as shown in Sect. 1.7.

Finally, introducing (4.18) in relation (4.17) leads to

τ

dξ

dt

= −(ξ −ξ

eq

) (t > 0), (4.19)

whose solution is an exponentially decreasing function

ξ −ξ

eq

= (ξ

0

−ξ

eq

) exp(−t/τ), (4.20)

where ξ

0

is the initial value of the degree of advancement. Within the limit

of an inﬁnite time, the system tends of course to its equilibrium state.

96 4 Chemical Reactions and Molecular Machines

Equation (4.20) is particularly interesting, because it provides some general

information about the relaxational process without a detailed knowledge of

the kinetics. If the stability condition (∂//∂ξ)

eq

< 0 is not satisﬁed, the

relaxation time will be negative, indicating that the degree of advancement

will not relax to equilibrium, but will increase exponentially.

4.2 Coupled Chemical Reactions

The study of coupled reactions is not merely a formal generalization of the

above considerations, it is also the basis of discussion of various relevant

concepts. For instance, if two processes are coupled, one of them may proceed

in the opposite direction it would have if it were alone. We will stress this

important feature in the next sections. Another interesting application of the

coupling between chemical reactions is provided by the triangular scheme

A B C A which will serve as a basis to establish Onsager’s reciprocal

relations.

4.2.1 General Formalism

The main modiﬁcations with respect to the results of Sect. 4.1 are the expres-

sions of the mass balance and the entropy production. Indeed, the variation

of the mass fractions c

k

of a given species is no longer due to one single

reaction, but depends on the several chemical reactions in which the given

species participates; it is now expressed as

dc

k

dt

=

r

j=1

ν

kj

M

k

dξ

j

dt

, k = 1, 2, . . ., n, (4.21)

where subscript j refers to the r diﬀerent chemical reactions, each of which

has its own degree of advancement ξ

j

, and ν

kj

is the stoichiometric coeﬃcient

of substance k in reaction j. By following the same steps as in Sect. 4.1, it is

found that, under isothermal conditions, the entropy production is

σ

s

=

j

/

j

T

w

j

. (4.22)

It is worth noting that the second law imposes that the whole sum in (4.22),

but not that each individual term, must be positive. Some terms may indeed

be negative, provided that their sum remains positive. Within the assumption

of linear ﬂux–force relations, one has

w

i

=

j

L

ij

/

j

T

, (4.23)

4.2 Coupled Chemical Reactions 97

with, in virtue of Onsager’s reciprocal relations,

L

ij

= L

ji

. (4.24)

In Sect. 4.2.2, a particular derivation of these symmetry relations for a trian-

gular reaction scheme is presented.

4.2.2 Cyclical Chemical Reactions and Onsager’s

Reciprocal Relations

A demonstration of the reciprocal relations (4.24), based on the cycle of

chemical reactions A B C A depicted in Fig. 4.1, was proposed by

Onsager in his celebrated papers of 1931. In Sect. 4.4, we will use a modiﬁed

form of this scheme as a model for chemically driven molecular motors.

Let k

i

(i = 1, 2, 3) be the kinetic constants, w

i

the respective velocities of

the reactions, which, according to the mass action law, are given by w

1

=

k

1

˜ c

A

−k

−1

˜ c

B

, w

2

= k

2

˜ c

B

−k

−2

˜ c

C

, w

3

= k

3

˜ c

C

−k

−3

˜ c

A

, and /

i

the respective

aﬃnities, i.e. /

1

= µ

A

− µ

B

, /

2

= µ

B

− µ

C

, /

3

= µ

C

− µ

A

; for simplicity,

all the stoichiometric coeﬃcients are supposed to be equal to one. Since the

process is cyclic, only two reactions are independent and /

1

+/

2

+/

3

= 0.

It will be shown that, when the ﬂux–force relations are of the form

w

1

−w

3

= L

11

/

1

+L

12

/

2

, (4.25a)

w

2

−w

3

= L

21

/

1

+L

22

/

2

, (4.25b)

the Onsager’s reciprocal relation L

12

= L

21

is automatically satisﬁed near

equilibrium, if the principle of detailed balance is valid. This principle states

Fig. 4.1 Triangular reaction scheme

98 4 Chemical Reactions and Molecular Machines

that transition A to B has the same probability as transition B to A, from

which it follows that, in equilibrium, w

1

= w

2

= w

3

= 0. Note that this

assumption does not follow directly from the phenomenological equations,

since non-zero but equal values w

1

= w

2

= w

3

,= 0 are compatible with

/

1

= /

2

= /

3

= 0. The mathematical details of the proof that L

12

= L

21

are found in Box 4.1.

Box 4.1 Mathematical Details of Onsager’s Demonstration

In virtue of (4.7c), the aﬃnities /

1

and /

2

may be written as

/

1

= RT ln

_

˜ c

A

(˜ c

A

)

eq

(˜ c

B

)

eq

˜ c

B

_

, /

2

= RT ln

_

˜ c

B

(˜ c

B

)

eq

(˜ c

C

)

eq

˜ c

C

_

, (4.1.1)

where use has been made of the results K

1

(T, p) = (˜ c

B

)

eq

/(˜ c

A

)

eq

and

K

2

(T, p) = (˜ c

C

)

eq

/(˜ c

B

)

eq

. By assuming that ˜ c

i

= (˜ c

i

)

eq

+ δ˜ c

i

, with

δ˜ c

i

¸ (˜ c

i

)

eq

, the aﬃnities may be expressed up to the ﬁrst order in the

deviations δ˜ c

i

from equilibrium as

/

1

= RT

_

δ˜ c

A

(˜ c

A

)

eq

−

δ˜ c

B

(˜ c

B

)

eq

_

, /

2

= RT

_

δ˜ c

B

(˜ c

B

)

eq

−

δ˜ c

C

(˜ c

C

)

eq

_

, (4.1.2)

and, from /

3

= −(/

1

+/

2

),

/

3

= RT

_

δ˜ c

C

(˜ c

C

)

eq

−

δ˜ c

A

(˜ c

A

)

eq

_

. (4.1.3)

Furthermore, according to the mass action law, one has

w

1

−w

3

= (k

1

+k

−3

)˜ c

A

−k

−1

˜ c

B

−k

3

˜ c

C

, (4.1.4)

w

2

−w

3

= k

2

˜ c

B

+k

−3

˜ c

A

−(k

−2

+k

3

)˜ c

C

. (4.1.5)

We now take into consideration the restrictions coming from the detailed

balance. From w

1

= w

2

= w

3

= 0, it is inferred that

k

1

(˜ c

A

)

eq

= k

−1

(˜ c

B

)

eq

, k

2

(˜ c

B

)

eq

= k

−1

(˜ c

C

)

eq

, k

3

(˜ c

C

)

eq

= k

−3

(˜ c

A

)

eq

.

(4.1.6)

Using (4.1.6) to eliminate k

−1

, k

−2

, and k

−3

in (4.1.4) and (4.1.5), one

obtains

w

1

−w

3

= k

1

(˜ c

A

)

eq

_

δ˜ c

A

(˜ c

A

)

eq

−

δ˜ c

B

(˜ c

B

)

eq

_

+k

3

(˜ c

C

)

eq

_

δ˜ c

A

(˜ c

A

)

eq

−

δ˜ c

C

(˜ c

C

)

eq

_

,

(4.1.7)

w

2

−w

3

= k

2

(˜ c

B

)

eq

_

δ˜ c

B

(˜ c

B

)

eq

−

δ˜ c

C

(˜ c

C

)

eq

_

+k

3

(˜ c

C

)

eq

_

δ˜ c

A

(˜ c

A

)

eq

−

δ˜ c

C

(˜ c

C

)

eq

_

.

(4.1.8)

4.2 Coupled Chemical Reactions 99

In view of the results (4.1.2) and (4.1.3), the above equations may

be written as

w

1

−w

3

=

k

1

RT

(˜ c

A

)

eq

/

1

−

k

3

RT

(˜ c

C

)

eq

/

3

=

1

RT

[k

1

(˜ c

A

)

eq

+k

3

(˜ c

C

)

eq

]/

1

+

k

3

RT

(˜ c

C

)

eq

/

2

, (4.1.9)

w

2

−w

3

=

k

2

RT

(˜ c

B

)

eq

/

2

−

k

3

RT

(˜ c

C

)

eq

/

3

=

k

3

RT

(˜ c

C

)

eq

/

1

+

1

RT

[k

2

(˜ c

B

)

eq

+k

3

(˜ c

C

)

eq

]/

2

, (4.1.10)

where use has been made of /

3

= −(/

1

+ /

2

). Note that (4.1.9) and

(4.1.10) have been derived exclusively by reference to chemical kinetics and

the principle of detailed balance.

A look on the coeﬃcients of /

2

in (4.1.9) and /

1

in (4.1.10), which referring

to (4.25a) and (4.25b), can be identiﬁed as L

12

and L

21

, respectively, leads

to the conclusion that

L

12

= L

21

. (4.26)

Let us now determine the corresponding results from classical irreversible

thermodynamics. The entropy production σ

s

is given by

Tσ

s

= w

1

/

1

+w

2

/

2

+w

3

/

3

, (4.27)

or, in terms of independent aﬃnities,

Tσ

s

= (w

1

−w

3

)/

1

+ (w

2

−w

3

)/

2

, (4.28)

which suggests the following constitutive relations:

w

1

−w

3

= L

11

/

1

+L

12

/

2

, (4.29)

w

2

−w

3

= L

21

/

1

+L

22

/

2

. (4.30)

Comparison with relations (4.1.9) and (4.1.10) yields

L

12

= L

21

=

k

3

RT

(˜ c

C

)

eq

, (4.31)

which completes the proof that the reciprocal relations are equivalent to the

principle of detailed balance.

It should be realized that these results are only valid near equilibrium,

where the net ﬂux along the cycle is zero. In molecular machines, the above

results are not directly transposable because the net ﬂux along the cycle is

diﬀerent from zero.

100 4 Chemical Reactions and Molecular Machines

4.3 Eﬃciency of Energy Transfer

A particularly appealing problem in non-equilibrium thermodynamics is to

determine the eﬃciency with which free energy is exchanged between coupled

chemical reactions. Indeed, in many biological situations, the transfer of free

energy (or free-energy transduction, as is often referred to in textbooks on

biochemistry) is of decisive importance. It is particularly true in microscopic

molecular motors – one of the frontiers of current research in biophysics –

which take free energy from some process (for instance, a chemical reaction)

and convert it into work (for instance, to carry out another chemical reaction

or a transport process). Note, however, that this subject cannot be studied

from the classical perspective of thermal engines, because biological systems

are often at homogeneous temperature, so that Carnot’s eﬃciency would be

zero. In contrast, the essential ideas of such isothermal free-energy transfer

can be well interpreted in the framework of linear non-equilibrium thermo-

dynamics.

Consider two coupled reactions, for which the dissipated energy is

Tσ

s

= w

1

/

1

+w

2

/

2

. (4.32)

The corresponding phenomenological equations describing the kinetics of the

reactions are

w

1

= L

11

/

1

+L

12

/

2

, (4.33)

w

2

= L

21

/

1

+L

22

/

2

. (4.34)

The above results remain applicable when one of the processes, say 2, is a

transfer of matter as considered in Sect. 4.4; such a situation is frequent in

biological cells. To guarantee the positiveness of Tσ

s

, either both terms at the

right-hand side of (4.32) are positive or one of them, say w

2

/

2

, is negative

but its absolute value is smaller than the ﬁrst one, assumed to be positive.

As a consequence, process 2 will evolve in opposite direction of its natural

tendency, for instance mass transport from a region of low to high concentra-

tion. A relevant question is the eﬃciency of the energy transfer from the ﬁrst

to the second reaction. Such kind of coupling is frequently met in biology.

For instance, in oxidative phosphorylation, the free energy delivered from the

oxidation of a substrate will serve to the phosphorylation of ADP (adenosine

diphosphate) to give ATP (adenosine triphosphate), which is a basic mole-

cule for metabolic energy exchanges. How much energy is transferred from

oxidation to phosphorylation is an interesting problem or, stated more gen-

erally, what is the eﬃciency of energetic conversion (Katchalsky and Curran

1965; Jou and Llebot 1990; Westerhoﬀ and van Dam 1987; Criado-Sancho

and Casas-V´azquez 2004).

In the above problem, reaction 1 is supplying some free energy, a part

of which is taken up by reaction 2 to proceed against its usual direction,

whereas the remaining part is lost under the form of heat. The objective

4.3 Eﬃciency of Energy Transfer 101

is to obtain the maximum eﬃciency from such an energy transfer (or en-

ergy transduction). The problem is parallel to that of Carnot in the context

of thermal engines. In the latter, heat was partially converted into work

and the remaining was dissipated. Here, the energy delivered by one reac-

tion is partially taken up by the other reaction and the remaining part is

dissipated.

In the following, we suppose that the ﬂuxes w

1

and w

2

are both positive

and that the forces are of opposite signs: /

1

> 0 and /

2

< 0. The degree of

coupling of the two processes is conveniently quantiﬁed by the coeﬃcient q

deﬁned by

q ≡

L

12

(L

11

L

22

)

1/2

. (4.35)

From the requirement that the dissipated energy is positive deﬁnite, one has

L

11

L

22

≥ (L

12

)

2

, and it follows that q will take values between −1 and +1:

for q = 1, the processes are completely coupled; for q = 0, they are totally

uncoupled. When q < 0, the processes are not compatible in the sense that

process (1) with w

1

/

1

> 0 is unable to produce a process like (2) with a

positive ﬂux w

2

and a negative force /

2

.

In relation with coupled processes, two further quantities have been in-

troduced: the stoichiometric coeﬃcient w

2

/w

1

and the eﬃciency η of the

conversion deﬁned by

η ≡ −

w

2

/

2

w

1

/

1

. (4.36)

Introducing the quantities x ≡ /

2

//

1

, Z ≡ (L

22

/L

11

)

1/2

, and using (4.33)

and (4.34), it is easily checked that

η = −

Zx(q +Zx)

1 +qZx

, (4.37)

which is maximum, namely, dη/dx = 0, at

x

max

=

(1 −q

2

)

1/2

−1

qZ

, (4.38)

thus leading to

η

max

=

q

2

[1 + (1 −q

2

)

1/2

]

2

. (4.39)

This result is worth to be underlined: in some aspects, it can be compared

with the classical Carnot’s result for the maximum eﬃciency of heat en-

gines, as it puts out an upper bound for the eﬃciency of energy transfer

between both reactions or, in general terms, between two coupled processes.

In heat engines, Carnot’s maximum eﬃciency depends only on the values of

the temperature of the heat reservoirs; here, the result (4.39) has a similar

universality: it depends only on the coupling coeﬃcient but not on the indi-

vidual phenomenological coeﬃcients. It is observed that the maximum of η

102 4 Chemical Reactions and Molecular Machines

−1.0 −0.8 −0.6 −0.4 −0.2 0.0

0.0

0.2

0.4

0.6

0.8

1.0

1.00

0.99

0.95

0.90

0.80

0.70

η

Zx

0.60

Fig. 4.2 Eﬃciency η vs. the parameter Zx for several values of the coupling coeﬃ-

cient q

diminishes with decreasing values of q and vanishes for q = 0. This is logical

because, in absence of coupling, the eﬃciency of the energy transfer should

be zero: this means that all the free energy delivered by reaction 1 would be

dissipated into heat, so that reaction 2 cannot take place against its usual

direction.

In Fig. 4.2, the values of the eﬃciency η as a function of the quantity Zx

for several values of the coupling coeﬃcient are reported. For q = 1, the

maximum of η is 1; in virtue of (4.36), this situation corresponds to w

1

/

1

=

−w

2

/

2

and consequently to a zero entropy production (reversible process).

The above considerations provide a clearer insight on the nature and im-

portance of the various phenomenological coeﬃcients, and they may be ex-

tended to cover a wider domain of situations, as the thermoelectric generators

in Sect. 3.2 (Van den Broeck 2005).

4.4 Chemical Reactions and Mass Transport:

Molecular Machines

The general developments of Sect. 4.3 will be applied to the study of molecular

machines, mainly pumps across membranes and motors along ﬁlaments which

are brieﬂy introduced in Box 4.2. In classical non-equilibriumthermodynamics,

4.4 Chemical Reactions and Mass Transport: Molecular Machines 103

Box 4.2 More About Biological Molecular Engines

An illustration of coupling between chemical reactions and transport is the

transport of macromolecules or vesicles along ﬁlaments of microtubules by

kinesin or dinein, two well-known molecular motors; these processes con-

tribute also to the formation of other organized intracellular motions, for

example the separation of chromosomes during cell division (see, for in-

stance, Nelson 2004). These motors must overcome the viscous resistance,

and this is achieved by using the free energy of chemical reactions. Another

example of chemically driven motion is that of the actin/myosin system

in myoﬁbrils, to which is due the muscle motility. In this system, bundles

of myosin molecules are interleaved with actin ﬁlaments; upon activation,

myosin crawls along the actin ﬁbres, shortening the muscle ﬁbre, the process

being fuelled by the hydrolysis of ATP.

Most cells are characterized by membrane pumps for the active trans-

port of Na

+

(towards the exterior) and K

+

(towards the interior); this is

especially important in nerve cells, wherein the action potential is due to a

transfer of Na

+

and K

+

across the excited membrane of the axon of neu-

rons. Other kinds of pumps in biological cells are pumps of Ca

2+

which

are found in muscles, and also in the synapses, and pumps of H

+

, which

are important in oxidative phosphorylation in mitochondria and in photo-

synthesis in chloroplasts. On the other side, the ﬂow of H

+

across some

molecular engines may gear molecular motors, as for instance the ﬂagellar

motor of some bacteria, which convert the electrochemical potential jump

of protons into mechanical torque for swimming.

under linear and isotropic conditions, chemical reactions (scalar processes) are

not coupled with mass transport (a vector process). However, the presence

of anisotropy may introduce a coupling between such processes, a frequent

situation in biological problems, as in active transport of ions or molecules

across cellular or tissue membranes, or along lengthy microscopic ﬁlaments.

These kinds of transport are attributed to enzymes, which act as molec-

ular engines. The study of microscopic machines, natural or artiﬁcial, is at

the frontiers of nowadays science. In the past, the calculation of the max-

imum eﬃciency of heat engines has represented a decisive contribution to

equilibrium thermodynamics. Similarly, the determination of the eﬃciency

of energy transfer between coupled processes may be considered as one of the

main achievements of linear irreversible thermodynamics.

As an illustration, consider the active transport of sodium through a cel-

lular membrane. Cells at rest state have an internal electrical potential of

−70 mV as compared to the external potential, taken as reference (0 mV).

Since the internal concentration of sodium is low as compared to the exter-

nal concentration, sodium has a natural tendency to ﬂow towards the interior

of the cell. Now it is observed that the cell pumps out as much sodium as the

ingoing one but, of course, this transport cannot be spontaneous, because

it takes place against the natural tendency of matter to ﬂow from regions

104 4 Chemical Reactions and Molecular Machines

of higher to lower electrochemical potential. This transport is made at the

expenses of the energy supplied by the hydrolysis of ATP, the usual fuel in

biological organisms.

We are in presence of a coupling between a chemical reaction, characterized

by an aﬃnity / and a chemical ﬂux w, and a transport of matter; it is shown

in Sect. 2.7 on matter diﬀusion that the latter is described by a force, the

diﬀerence of chemical potential ∆µ between the inner and the outer regions,

and a ﬂux of matter (sodium in the present case) J. The corresponding rate

of dissipation is given by

Tσ

s

= /w +J∆µ, (4.40)

with /w > 0, because the hydrolysis of ATP takes place along the natural

tendency given by / > 0 and w > 0, whereas J∆µ < 0, because the ﬂow

of sodium (J < 0) is opposite to the direction of ∆µ. We are faced with

an example where a reaction with positive entropy production allows for the

occurrence of a simultaneous process, with a negative contribution to the

entropy production.

The phenomenological laws describing the kinetics of the reaction and the

rate of transport are given by

w = L

11

/+L

12

∆µ, (4.41)

J = L

21

/+L

22

∆µ, (4.42)

with L

12

= L

21

according to Onsager’s reciprocal relations. The coeﬃcient

L

21

in (4.42) is responsible for the coupling between chemical reaction and

mass transport.

For instance, the experimental values of the phenomenological coeﬃcients

for the transport across the membrane of a frog bladder are the following:

L

22

= 104, L

12

= 5.40, and L

11

= 0.37 in units of mol

2

cm

−2

s

−1

kcal

−1

. The

coupling parameter q deﬁned by (4.35) is given by q = 0.86 and the maximum

eﬃciency calculated from (4.39) is η

max

= 33%.

In so-called linear motors, i.e. motors which transport matter along lengthy

linear macromolecules, as kinesin along microtubules, the resistance is due

to viscous friction rather than to a jump of the electrochemical potential. In

Box 4.3, the modus operandi of such a molecular motor is presented. The

expression for the dissipated energy is then

Tσ

s

= /w +vF, (4.43)

v being the velocity of the machine along the ﬁlament and F the viscous

resistance. The phenomenological laws are, in the linear approximation,

w = L

11

/+L

12

F, (4.44)

v = L

21

/+L

22

F, (4.45)

4.4 Chemical Reactions and Mass Transport: Molecular Machines 105

with the coeﬃcient L

22

inversely proportional to the viscosity of the medium.

Since molecular machines are of very small dimensions, the energy related to

their several steps is comparable or smaller than the thermal noise. Their

behaviour is therefore not deterministic but stochastic, in contrast with that

of macroscopic engines. The understanding of the relations between thermal

ﬂuctuations and the behaviour of molecular machines is a subject of current

interest, as explained in Box 4.4.

Box 4.3 Biological Molecular Motors with Three Conﬁgurations

A model for chemically driven transport may be based on a modiﬁcation

of the triangular reaction scheme of Sect. 4.2.2. An abstract illustration is

provided by the following molecular motor running through three diﬀerent

conﬁgurations, denotedas M

1

, M

2

, andM

3

(there wouldbe no diﬃcultyto add

more conﬁgurations), and working in a cyclic way. We assume that the motor

is fuelled with ATP and hydrolyses it, changing the conﬁguration at each

chemical step. Thus, instead of a closed reaction triangle, one has the scheme

(M

1

) + ATP →(M

2

) ATP, (4.3.1)

(M

2

) ATP →(M

3

) ADP + P

i

, (4.3.2)

(M

3

) ADP →(M

1

) + ADP. (4.3.3)

To be explicit, conﬁguration 1 bounds one molecule of ATP, which pro-

duces a change to conﬁguration M

2

. In a second step, ATP is hydrolysed:

the inorganic phosphate P

i

is liberated and the macromolecule changes to

conﬁguration M

3

, with ADP still bound to it. Finally ADP is liberated

from conﬁguration M

3

and the motor returns to the initial conﬁguration

M

1

, where the cycle may start again. At each cycle, the motor does some

work: either it advances a step along a macromolecule (as for instance ki-

nesin along tubulin or myosin along actin) or it translocates some molecule

(as for instance in membrane ionic pumps). For more biological details, the

reader is referred, for instance, to Nelson (2004).

Finally, the triangular scheme would be

106 4 Chemical Reactions and Molecular Machines

It is assumed that the kinetic rate constants of reactions (4.3.1)–(4.3.3) and

k

−1

, k

2

, and k

3

in the triangular scheme do not depend on the ATP nor

ADP concentration, whereas k

1

= k

(0)

1

[ATP], k

−3

= k

(0)

−3

[ADP], and k

−2

=

k

(0)

−2

[P] (where k

(0)

1

, k

(0)

−3

, and k

(0)

−2

are constant values) are pseudo-ﬁrst-order

rate coeﬃcients into which the concentrations of ATP, ADP, and P

i

are

incorporated. This makes an essential diﬀerence with the triangular reaction

scheme presented in Sect. 4.2.2. Indeed when the hydrolysis reaction

ATP →ADP + P

i

(4.3.4)

is in equilibrium, the cycle has not net motion; however, when the aﬃnity

of this reaction is positive, the cycle will turn in the direction 123, whereas

if the aﬃnity is negative, it will move in the opposite direction 321. Such

kind of model machine may both hydrolyse ATP and produce some work

pumping ions from lower to higher electrochemical potential (active trans-

port) or, in the reverse way, it may draw some energy from the ions crossing

it from higher to lower electrochemical potential and produce ATP.

Box 4.4 Brownian Motors

Molecular motors are sensitive to thermal noise which might be used, to

some extent, to produce useful work, provided some structural and ther-

modynamic conditions are satisﬁed. Instead of performing a simple random

Brownian motion, these motors may rectify, to some extent, the thermal

noise. The modelling of such molecular motors has been much developed

since 1990 (Magnasco 1994; Astumian and Bier 1994; J¨ ulicher et al. 1997;

Astumian 1997) (Fig. 4.3).

Fig. 4.3 Transport driven by a chemical reaction (Astumian 1997)

4.4 Chemical Reactions and Mass Transport: Molecular Machines 107

As an example, we analyse a simple model proposed by Astumian (1997)

for chemically driven transport. The author studies the one-dimensional

motion of a charged particle along a periodic lattice of dipoles arranged

head to tail (as found in many biological macromolecules, like tubulin or

actin). The interaction between the particle and the linear arrangement

of dipoles is assumed to be merely electrostatic (in biology, the couplings

may also be conformational). The potential energy proﬁle has essentially

the form of a sawtoothed function as that shown in Fig. 4.3 (if the particle

is negatively charged, the minima of the sawtooth will correspond to the

positive ends of the dipoles and the maxima to the negative ends). Because

of thermal noise, the particle undergoes Brownian motion along the line,

and occasionally has enough energy to pass one of the neighbouring barriers.

However, despite the anisotropy of the potential proﬁle, the probabilities of

moving to the right or to the left are equal.

Assume, instead, that the particle catalyses a chemical reaction SH

S

−

+H

+

, with the product particles being charged; S designates some sub-

stract species. Now, the amplitude of the potential will depend on whether

H

+

or S

−

are bound to the particle, resulting in a coupling between the

chemical reaction and the diﬀusion of the particle along the line. However,

when the particle is bound only to H

+

(assume that the ionized substract

S

−

has been expelled to the bulk), the particle becomes uncharged and there

is no longer a barrier opposing its motion. Therefore, the particle may move

along the line. Once H

+

is liberated, the charge of the particle constitutes

again a barrier. However, it may be shown that an asymmetric potential

whose barriers are ﬂuctuating in height gives raise to a directional motion

of particles. The corresponding ﬂux is given by (Astumian and Bier 1994)

J ≈

γk

B

T

4

_

erf

_

α

2

ζ

_

−erf

_

1 −α

2

ζ

__

,

where α is the asymmetry parameter of the potential, and ζ ≡ (γL

2

/D)

1/2

while γ, L, and D are typical parameters of the model. If the saw-

toothed potential is symmetric (α = 0.5), J = 0 and there is no net

ﬂow.

To summarize, it turns out that molecular machines can convert free en-

ergy into motion under the conditions that this mechanism is out of equi-

librium and structurally asymmetric. Moreover, molecular engines cannot

increase their speed without limit as the energy supply is increased, because

the speed saturates at some limiting value. This means that linear transport

relations are not indeﬁnitely valid, but break down for high values of ﬂuxes

and forces.

108 4 Chemical Reactions and Molecular Machines

4.5 Autocatalytic Reactions and Diﬀusion:

Morphogenesis

Another consequence of the coupling between chemical reactions and diﬀu-

sion is the occurrence of patterns, i.e. spatial inhomogeneities, in systems far

from equilibrium. Although this topic is studied in detail in Chap. 6, we ﬁnd

it interesting to present here an introductory version based on a model pro-

posed by Turing (1952), who settled the foundations of morphogenesis, i.e. the

transition of an originally homogeneous system into an inhomogeneous one.

This is a characteristic feature of non-equilibrium thermodynamics, because

near equilibrium, transport phenomena (matter diﬀusion, thermal conduc-

tion) tend to make the system spatially homogeneous in the long run.

Consider two ﬂuid subsystems at the same pressure and temperature,

which are separated by a permeable membrane. In each of them, the fol-

lowing autocatalytic reaction will take place between substances A and X

A + 2X →3X. (4.46)

In this reaction, a substance A is transformed into the substance X, at a rate

which depends on the concentration of X. The molar concentrations of X

in both subsystems, ˜ c

X

1

and ˜ c

X

2

, respectively, tend to become equal due to

diﬀusion of X across the membrane. The diﬀusion ﬂux of X from subsystem

1 to subsystem 2 is given by

J

1→2

= α(˜ c

X

1

− ˜ c

X

2

), (4.47)

where α is a constant proportional to the permeability and the area of the

membrane.

We will show that, under some conditions, the diﬀerence of concentrations

˜ c

X

between both subsystems will increase because of the autocatalytic re-

action (4.46). First, we note that the rate of production of substance X is

higher where the molar concentration ˜ c

X

is higher, as it follows from the mass

action law

d˜ c

X

dt

= k˜ c

A

(˜ c

X

)

2

, (4.48)

where k is a positive kinetic constant.

Taking diﬀusion into account, the evolution of ˜ c

X

1

as a function of time in

subsystem 1 is given by

d˜ c

X

1

dt

= −α(˜ c

X

1

− ˜ c

X

2

) +k˜ c

A

(˜ c

X

1

)

2

, (4.49)

whereas the evolution of ˜ c

X

2

in subsystem 2 is

d˜ c

X

2

dt

= −α(˜ c

X

1

− ˜ c

X

2

) +k˜ c

A

(˜ c

X

2

)

2

. (4.50)

By subtracting both equations, one obtains for the evolution of the diﬀerence

of concentrations ˜ c

X

1

− ˜ c

X

2

,

4.6 Problems 109

d(˜ c

X

1

− ˜ c

X

2

)

dt

= −2α(˜ c

X

1

− ˜ c

X

2

) +β˜ c

A

_

(˜ c

X

1

)

2

−(˜ c

X

2

)

2

¸

. (4.51)

This may still be written as

d(˜ c

X

1

− ˜ c

X

2

)

dt

= −2α(˜ c

X

1

− ˜ c

X

2

) +k˜ c

A

(˜ c

X

1

+ ˜ c

X

2

)(˜ c

X

1

− ˜ c

X

2

). (4.52)

This result indicates that the right-hand side of (4.52) becomes positive,

thus implying an increase of the diﬀerence of concentrations, when the molar

concentration of A becomes higher than the critical value

(˜ c

A

)

c

=

2α

k

(˜ c

X

1

+ ˜ c

X

2

). (4.53)

Note that, since reaction (4.46) is irreversible, i.e. it only goes to the right, the

concentration of A corresponding to equilibrium is zero. In the problem under

study, ˜ c

A

remains constant because it is supplied from the outside at the same

rate in both subsystems. If the supply rate is suﬃciently high in such a way

that the concentration of A exceeds the critical value, the concentration ˜ c

X

will tend to be higher in one subsystem than in the other and will contribute

to the reinforcement of the non-homogeneity of the system.

This mechanism has gained much interest in biology. Indeed, an actual

problem in biology is the development of an embryo from the ﬁrst fertilized

cell of the individual and this process implies cell diﬀerentiation at the var-

ious stages of development. Though all the cells of the individual contain

the same information in their DNA, the parts of this information which are

really read are not the same in all cells, but they depend on control proteins

that determine which genes are transduced into the corresponding proteins.

In the above model, one could interpret X as the control molecule determining

the reading of some gene. When the system is far enough from equilibrium,

the inhomogeneity in the value of ˜ c

X

makes that, in one part of the organism

where ˜ c

X

is higher than some critical value, a given gene is expressed, whereas

in the other regions, where ˜ c

X

is lower, the gene is not expressed.

4.6 Problems

4.1. Michaelis–Menten’s relation for enzymatic reactions. An important type

of biochemical reactions are reactions changing a substrate S into a product P,

catalysed by an enzyme E, through the intermediate formation of an enzyme–

substrate complex ES. The reaction may be expressed as

E + S ES →E + P.

The ﬁrst step is reversible and the second one is irreversible. The correspond-

ing kinetic constants are k

1

and k

−1

for the forward and backward steps of

110 4 Chemical Reactions and Molecular Machines

the ﬁrst reaction, and k

2

for the (forward) step of the second reaction. The

velocity of reaction w ≡ d˜ c

P

/dt = −d˜ c

S

/dt obeys the Michaelis–Menten

expression

w = w

max

˜ c

S

K

M

+ ˜ c

S

,

where w

max

is the saturation velocity and K

M

a characteristic constant indi-

cating the value of the substrate molar concentration ˜ c

S

for which the velocity

is half the maximum value. (a) Show that, according to the mass action law,

the value of the concentration of the complex ES in the steady state (for

which the enzyme concentration ˜ c

E

is constant) is given by

˜ c

ES

=

k

1

˜ c

S

k

−1

+k

2

+k

1

˜ c

S

.

(b) It follows from the mass action law that the velocity of the reaction is

w = d˜ c

P

/dt = k

2

˜ c

ES

. Combine this result with the previous one and show

that Michaelis–Menten expression is recovered. Identify w

max

and K

M

in

terms of the kinetic constants k

1

, k

−1

, k

2

, and of the enzyme concentration

˜ c

E

. Discuss the dependence of w

max

with ˜ c

E

and the relevance of the fact

that the last step is totally irreversible (i.e. the backward reaction is lacking).

(c) Assume that for a given enzymatic reaction w

max

= 0.085 mMs

−1

and

K

M

= 6.5 mM. Obtain the velocity w for ˜ c

S

= 2.5, 10, and 25 mM.

4.2. Application of Michaelis–Menten expression to simple molecular motors.

The kinetics of simple, tightly coupled chemical motors with at least one ir-

reversible step is expected to be described by a Michaelis–Menten expression,

discussed in the previous problem. Assume, for instance, the following scheme

motor(x) + ATP ATP motor →motor(x + ∆x) + ADP.

In this reaction, the hydrolysis of ATP gives the energy necessary for moving

the motor from a position x to x +∆x along some ﬁlament, the motor could

be kinesin and the corresponding ﬁlament could be a microtubule, in which

case ∆x = 8 nm. The speed with which the motor moves along the ﬁlament

can be cast in the form

v = v

max

(F)

˜ c

ATP

K

M

(F) + ˜ c

ATP

,

wherein the maximum velocity v

max

and the characteristic constant K

M

de-

pend on the load force acting on the motor. Indeed, it has been found that,

for kinesin along microtubules, the numerical values are v

max

≈ 815 nms

−1

and K

M

≈ 90 µM at F = 1 pN, and v

max

≈ 400 nms

−1

and K

M

≈ 300 µM at

F = 5.5 pN. (a) Compare the velocity at ˜ c

ATP

= 1, 10, and 100 mM, under

both load charges. (b) Compare the useful power delivered by the motor,

given by Fv, in the several situations mentioned above.

4.3. Conditions of maximum eﬃciency. Consider a transport process across

a membrane, when the aﬃnity of the hydrolysis of ATP is /

1

and assume

4.6 Problems 111

that the transported substance may be considered as ideal and uncharged.

Show that the ratio of internal and external concentrations of the substance,

under the conditions of optimum free-energy exchange between ATP energy

and transfer, is given by

ln

c

in

c

out

=

/

1

RT

(1 −q

2

)

1/2

−1

qZ

,

the coupling coeﬃcients q and Z are deﬁned in Sect. 4.3, namely q ≡

L

12

/(L

11

L

22

)

1/2

and Z ≡ (L

11

/L

22

)

1/2

, L

ij

being the phenomenological

coeﬃcients of the linear constitutive equations relating the velocity of the

chemical reaction (taken as ﬂux 1) and the rate of active transport (taken as

ﬂux 2) with the corresponding conjugated thermodynamic forces.

4.4. Kinetics of triangular reaction. An example of a triangular reaction

scheme is provided by the triangular catalytic isomerization of the three

butenes C

3

H

8

(namely, forms (A), (B), and (C) shown below), was stud-

ied by W. O. Haag and H. Pines (J. Am. Chem. Soc. 82 (1960) 387 and

2488). They obtained for the corresponding kinetic constants in the presence

of Al

2

O

3

catalyst: k

1

/k

3

= 2.4, k

−2

/k

−1

= 1.0, and k

−3

/k

2

= 0.4. Analyse

whether these results are compatible with the statement that the global equi-

librium in the triangular reactions corresponds to equilibrium in each of the

three individual reactions (see Chartier et al. 1975).

(A) H

2

C == CH — CH

2

— CH

3

( ) B CH — CH

3

(cis)

CH — CH

3

( ) CH — CH

3

C

CH

3

— CH

(trans)

4.5. Autocatalytic reactions and diﬀusion. Discuss the instability problem

analysed in Sect. 4.5 when one assumes that A+X →2X instead of A+2X →

3X, as in (4.46).

4.6. Entropy production and degree of coupling in kinesin motor. Kinesin, a

typical molecular machine which moves along microtubules, takes on average

one 8-nm step, corresponding to one monomer of tubulin, every 10 ms. A

single ATP molecule is hydrolyzed per step. The chemical energy relased by

ATP hydrolysis is typically 20 k

B

T, and the motor uses 12 k

B

T with each step

(Bustamante et al. 2005). a) Compute the power and the force done by the

kinesin, and the hydrolysis velocity reaction. b) What is the eﬃciency of the

112 4 Chemical Reactions and Molecular Machines

machine? c) Evaluate the entropy production of the machine in terms of k

B

per second. d) If this eﬃciency is the maximum one, evaluate from equation

(4.39) the degree of coupling between ATP hydrolysis and kinesin motion,

according to the scheme provided in (4.44) and (4.45) for linear motors.

4.7. Bacterial ﬂagellar motor. Swimming bacteria are usually propelled by

a rotary motor embedded in the cell envelope which drives rotating ﬂagellar

ﬁlaments. These motors are typically powered by the inﬂow of protons across

the cell membrane, and have very interesting properties which confer bacteria

with the possibility of chemiotaxis, i.e., swimming towards higher nutrient

concentrations. We simplify the complexity of this motor and assume that

the power delivered by the protons, namely Power = J∆µ, is applied with

unit eﬃciency to propel the cell. Here J is the proton ﬂux across the motor

(number of protons per second) and ∆µ is the socalled protonmotive force,

deﬁned as

∆µ = ∆V + 2.303(k

B

T/e)∆pH

∆V being the membrane potential (diﬀerence of potential across the mem-

brane) and ∆pH being ∆pH = log(c

H+,out

/c

H+,in

), with c

H+

being the proton

concentration. Assume that a bacterium is a sphere of 1 µm of radius, and

its friction coeﬃcient is given by the Stokes law, i.e. ζ = 6πηr, with r the

radius of the cell and η the viscosity of the surrounding ﬂuid. Assume that

when the membrane potential is −80 mV (inside is negative) and ∆pH = 0,

the bacterium moves at 25 µm/s. a) Find the proton ﬂux necessary to propel

the cell in water (η = 10

−3

N s m

−2

). b) Find the electrical resistance of the

motor. c) Evaluate the velocity of the cell when it swims in a solution with

pH = 6 (assume that inside the cell pH = 7). d) Compare the total entropy

production due to viscous dissipation in both cases.

4.8. Degree of coupling and maximum power. In expression (4.39) was given

the maximum eﬃciency in terms of the degree of coupling q of two coupled

chemical reactions. In some circumstances, however, as for instance in hunting

or ﬂeeing, so important for survival, the organism is interested in obtaining

the maximum power instead of the maximum eﬃciency. a) Write the power

output −w

2

A

2

in the normalized form −w

2

A

2

/L

11

A

2

1

and show that it may

be expressed as −Zx(Zx + q); b) Show that the corresponding eﬃciency at

the value of q for which the power output is maximum is

η

max power

= (q

2

/4)[1(q

2

/2)]

−1

.

c) Compare this value with the maximum eﬃciency.

4.9. Eﬀects of the breaking of detailed balance on the Onsager relations. As-

sume that in the triangular cyclical set of chemical reactions presented in

Section 4.2.2, detailed balance was broken, i.e., instead of having w

1

= w

2

=

w

3

= 0 one had w

1

= w

2

= w

3

= w. How would the phenomenological coeﬃ-

cients L

12

and L

21

appearing in the constitutive equations (4.25) be related?

(i.e. express L

12

−L

21

in terms of w).

Chapter 5

Finite-Time Thermodynamics

Economy, Ecology, and Heat Engines

One of the motivations of the early developments of thermodynamics was

the optimization of heat engines, as steam engines, which transform partially

heat into work, and which led to the ﬁrst industrial revolution between the

end of the eighteenth century and the beginning of the nineteenth century.

This provides an interesting illustration of how an engineering problem cou-

pled with economical needs stimulated the occurrence of a new fundamental

science. Nowadays, we live in a similar situation: we are experiencing an in-

dustrial revolution based on information processing and miniaturized engines

and, on the other side, we are conscious of the need to incorporate ecological

restrictions into the economical progress. To convey some ﬂavour of these

problems, we present in this chapter applications of non-equilibrium ther-

modynamics to heat engines working with a ﬁnite non-vanishing rate. This

approach is referred to as ﬁnite-time thermodynamics and received a strong

impetus during the last quarter of the twentieth century.

In 1824, Carnot arrived to the conclusion that the eﬃciency of a heat

engine operating between a hot and a cold heat reservoir is maximum in

reversible processes, and that it depends only on the temperature of the

two reservoirs, but not on the working substance. Here, Carnot’s ideas are

extended by including considerations not only on the eﬃciency but also on

the power developed by the engine when the cycle is performed in a ﬁnite

time. This issue is of importance from the economical perspective, because

what the society needs is an amount of work in a ﬁnite time, rather than

work supplied at an inﬁnitesimal small rate, as in reversible engines.

Though the mutually conﬂicting demands in the maximization of eﬃciency

and power are especially illustrative of the conceptual problems dealt with

in this chapter, we also compare other optimization criteria, as for instance,

those based on the minimization of entropy production, and on maximization

of the power minus the product of temperature and entropy production.

This diversity of criteria illustrates the conceptual subtleties arising in ﬁnite-

time thermodynamics. Furthermore, the time constraints must be included

in realistic models of engines to optimize their design and performance.

113

114 5 Finite-Time Thermodynamics

Finite-time thermodynamics deals with a very wide collection of situa-

tions. In the present chapter, we only present some simple and pedagogical

illustrations, based on Carnot’s engine, but many other problems have been

investigated in ﬁnite-time thermodynamics as, for instance, how to optimize a

thermodynamic cycle when its total duration is ﬁxed. In this case, the several

branches of the cycle may be performed at diﬀerent rates, and the problem is

to ﬁnd the rates of each process in such a way that some optimization crite-

rion (power output maximization, for instance) is satisﬁed. Finite-time ther-

modynamics is not limited to Carnot cycle, but it encompasses many other

cycles (Otto, Brayton, Stirling, Diesel, etc.), and it is concerned not only

with ideal gases but also with other kinds of ﬂuids, as polytropic ﬂuids. Par-

ticular studies have been devoted to refrigerators, heat pumps, engines with

heat regeneration, several coupled engines, and have been applied to solar

engines, photovoltaic cells, or to the evaluation of the maximum eﬃciency of

the conversion from solar energy to wind energy in the planet. In short, ﬁnite-

time thermodynamics allows us to understand in a simple way the essential

role of time constraints on the thermodynamic cycles (Salamon et al. 1980;

Ma 1985; Salamon and Sieniutycz 1991; Bejan 1996; Hoﬀmann et al. 1997).

This is a clear illustration of the importance of including non-equilibrium

thermodynamics both in the assessment and design of heat engines as well

as in the understanding of some natural phenomena.

5.1 The Finite-Time Carnot Cycle

Let us brieﬂy recall the essentials of the classical analysis of Carnot’s engine.

The latter is operating between two heat reservoirs at temperatures T

1

and T

2

(T

1

> T

2

), it takes an amount of heat Q

1

from the hot one, transforms a part

of it into work W, and delivers the remaining heat Q

2

to the cold reservoir.

Since the engine operates cyclically, Q

1

, Q

2

, and W refer, respectively, to the

amounts of heat, absorbed and delivered, and the work carried out by the

engine during one cycle.

The internal energy of the engine does not change during one cycle, because

it is a state function, so that the ﬁrst law takes the form

Q

1

= W +Q

2

. (5.1)

If we ask about the eﬃciency η of the cycle – namely the work produced W

divided by the heat supplied Q

1

to the engine – one has

η ≡

W

Q

1

= 1 −

Q

2

Q

1

. (5.2)

Kelvin–Planck formulation of the second law states the impossibility of build-

ing a cyclical heat engine, which transforms heat completely into work. In-

deed, the maximum value of η is limited by the minimum value of the heat

5.1 The Finite-Time Carnot Cycle 115

Q

2

delivered to the cold reservoir. Therefore, to evaluate the maximum eﬃ-

ciency, the second law must be taken into account. Accordingly, the global

balance of entropy is never negative, i.e.

∆S

tot

= ∆S

1

+ ∆S

2

+ ∆S

engine

≥ 0, (5.3)

where ∆S

1

, ∆S

2

, and ∆S

engine

are the variations of entropy, respectively, in

the reservoirs 1 and 2 and in the engine during one cycle. Since the engine

operates cyclically, and recalling that the entropy is a function of state, one

has ∆S

engine

= 0. Moreover, the temperature of each reservoir remains con-

stant, so that, according to Clausius’ deﬁnition of entropy, the variations of

their respective entropies are ∆S

1

= −Q

1

/T

1

and ∆S

2

= Q

2

/T

2

. Therefore,

(5.3) becomes

∆S

tot

= −

Q

1

T

1

+

Q

2

T

2

= ∆

irr

S ≥ 0. (5.4)

The equality sign corresponds to reversible processes and ∆

irr

S stands for

the entropy produced during one cycle by irreversible processes (as for in-

stance thermal resistance or mechanical and hydrodynamical friction). After

combining (5.2) and (5.4), it follows that the eﬃciency obeys the inequality

η ≤ 1 −

T

2

T

1

. (5.5)

This expression shows that the eﬃciency is maximum when the engine op-

erates reversibly: indeed, in this case the relation (5.5) becomes an equality,

providing the so-called Carnot’s eﬃciency, depending only on the temper-

ature of the sources but not on any detail of the working substance or the

engine itself.

However, in order that the engine operates reversibly, one must proceed

inﬁnitely slowly, with the result that the power delivered by the reversible

engine is in fact zero. This is of no practical interest because, in actual sit-

uations, a non-vanishing power is desirable, which implies to obtain a ﬁnite

amount of work in a ﬁnite time. Anyway, one should avoid spending unre-

alistic eﬀorts to improve engines beyond their actual physical limits; in that

respect, the result (5.5) indicates that a way to enhance the eﬃciency is to

raise as much as possible the temperature of the hot reservoir and to reduce

the temperature of the cold reservoir.

A cycle carried out in a ﬁnite time means a reduction of the eﬃciency, and

therefore one is faced with an alternative: maximization of eﬃciency or max-

imization of power. To study this problem, we will follow in this introductory

section the didactical example proposed by Curzon and Ahlborn (1975).

5.1.1 Curzon–Ahlborn’s Model: Heat Losses

Curzon and Ahlborn assume that the rate of heat exchanged between the

working substance at a temperature T

1

and the heat reservoir at a tempera-

ture T

1

is described by Newton’s law of heat transport, namely

116 5 Finite-Time Thermodynamics

Fig. 5.1 Sketch of the Curzon–Ahlborn’s model of heat engine with ﬁnite thermal

resistance between the working ﬂuid and the heat reservoirs

dQ

1

dt

= α

1

(T

1

−T

1

), (5.6)

where the coeﬃcient α

1

depends on the thermal conductivity of the wall

separating the working ﬂuid from the source and on geometrical factors, as

the area and the width of the wall. Analogously, the rate of heat delivered to

the cold source is

dQ

2

dt

= α

2

(T

2

−T

2

), (5.7)

where α

2

plays a role analogous to α

1

in the heat exchange between the

working ﬂuid and the cold reservoir. A sketch of the Curzon–Ahlborn’s model

is provided in Fig. 5.1.

It follows from (5.6) and (5.7) that the times required for the engine to

absorb an amount of heat Q

1

and to deliver the heat Q

2

are, respectively,

given by

t

1

=

Q

1

α

1

(T

1

−T

1

)

, t

2

=

Q

2

α

2

(T

2

−T

2

)

. (5.8)

By writing these expressions, we have assumed that the temperatures T

1

and T

2

of the working substance remain constant during the heat exchange

process. Furthermore, for the sake of simplicity, it is admitted that the time

spent in the adiabatic parts of the cycle is much shorter than the time required

for absorbing or delivering the heat. Physically, this implies the absence of

frictional losses and inertial eﬀects (e.g. one works with a frictionless piston

of zero mass). In this case, the total duration of the cycle is

t

cycle

≈ t

1

+t

2

. (5.9)

This theoretical framework has the advantage of allowing explicit calculations

of all the quantities related to the way the heat engine is operating, and to

5.1 The Finite-Time Carnot Cycle 117

introduce explicitly the notion of time. In comparison with reversible ther-

modynamics, one makes a step forward by capturing more realistic features

of the heat engine.

The power P delivered by the engine during one cycle is the ratio of the

work W that is performed and the total duration of the cycle, i.e.

P =

W

t

cycle

. (5.10)

Now, we want to obtain both eﬃciency and power as a function of the dura-

tion of the cycle. To do this, we need the expression of the entropy produced

during a cycle, which is given by (5.4). On the other hand, we know from

Chap. 2 that, when an amount of heat Q ﬂows from a system at temperature

T to another at temperature T

**, the amount of entropy produced is
**

∆

irr

S = −Q

_

1

T

−

1

T

_

. (5.11)

The total entropy generated during one cycle is assumed to result exclusively

from irreversible heat exchanges between the reservoirs and the working sub-

stance, all other possible sources of irreversibility, as for instance viscosity,

are neglected for the moment. In the above problem, one has

∆

irr

S = −Q

1

_

1

T

1

−

1

T

1

_

−Q

2

_

1

T

2

−

1

T

2

_

. (5.12)

Making use of (5.4), one obtains

−

Q

1

T

1

+

Q

2

T

2

= 0. (5.13)

Compared to the equivalent result (5.4), wherein the temperatures T

1

and T

2

of the sources appear, in (5.13) the temperatures of the working ﬂuid dur-

ing the isothermal expansion and compression are coming out. The analogy

between (5.13) and (5.4) for the reversible situation (i.e. with ∆

irr

S = 0)

underlines that the working ﬂuid itself is behaving in a reversible way, and

that the only source of irreversibility is the heat transfer between systems at

diﬀerent temperatures (i.e. between the working ﬂuid and the corresponding

heat reservoirs). This is called an endoreversible heat engine (de Vos 1992;

Andresen 1996; Hoﬀmann et al. 1997), because irreversibilities ﬁnd their ori-

gin in the contact with the sources but not in internal causes. In particular,

for a given compression ratio V

max

/V

min

of an ideal gas and using the ﬁrst

law of thermodynamics, the values of Q

1

and Q

2

are

Q

1

= RT

1

ln(V

max

/V

min

), Q

2

= RT

2

ln(V

max

/V

min

). (5.14)

We now study the conditions under which the power becomes a maximum.

The work per cycle W is equal to the diﬀerence Q

1

−Q

2

, and will be the larger

118 5 Finite-Time Thermodynamics

Fig. 5.2 Normalized eﬃciency and power output of the heat engine as a function

of the inverse of the duration of the cycle

when the higher is Q

1

and the lower is Q

2

. According to (5.13), this requires

to increase T

1

and to reduce T

2

, because Q

1

/Q

2

= T

1

/T

2

. But if T

1

increases

and T

2

decreases, the rate of heat exchanged with the sources, expressed by

(5.6) and (5.7), diminishes. Under these circumstances, one has slow cycles

with high eﬃciency but small power: each cycle produces much work but one

cycle takes a long time. If the rate of heat is increased by lowering T

1

and

rising T

2

, eﬃciency is reduced: each cycle produces less work but it takes a

shorter time. When the rate is too high, the work per cycle becomes very low

in such a way that the power decreases and becomes eventually zero. Between

these extreme situations with a vanishing power, there is another one with

a maximum power output, as sketched in Fig. 5.2. In this ﬁgure, the zero

values of power output correspond either to very low speed per cycle (high

eﬃciency but small power) or to very high speed per cycle (low eﬃciency and

low power).

To determine the values of T

1

and T

2

corresponding to the maximum power

output, at ﬁxed values of the temperatures T

1

and T

2

of the heat reservoirs,

let us write explicitly the expression for the power output, namely

P(T

1

, T

2

) =

Q

1

−Q

2

t

cycle

=

Q

1

−Q

2

Q

1

α

1

(T

1

−T

1

)

+

Q

2

α

2

(T

2

−T

2

)

. (5.15)

Eliminating Q

1

and Q

2

in terms of T

1

and T

2

by using (5.13), one obtains

P(T

1

, T

2

) =

T

1

−T

2

T

1

α

1

(T

1

−T

1

)

+

T

2

α

2

(T

2

−T

2

)

. (5.16)

After introducing the variables x

1

≡ T

1

− T

1

and x

2

≡ T

2

− T

2

, (5.16) may

be rewritten in the more compact form

5.1 The Finite-Time Carnot Cycle 119

P(x

1

, x

2

) =

α

1

α

2

x

1

x

2

(T

1

−T

2

−x

1

−x

2

)

α

2

T

1

x

2

+α

1

T

2

x

1

+ (α

1

−α

2

)x

1

x

2

. (5.17)

The conditions for P being extremum are

∂P(x

1

, x

2

)

∂x

1

= 0,

∂P(x

1

, x

2

)

∂x

2

= 0. (5.18)

This leads to two equations for the values of x

1

and x

2

corresponding to the

maximum power,

α

2

T

1

x

2

(T

1

−T

2

−x

1

−x

2

) = x

1

(α

2

T

1

x

2

+α

1

T

2

x

1

+ (α

1

−α

2

)x

1

x

2

), (5.19)

α

1

T

2

x

1

(T

1

−T

2

−x

1

−x

2

) = x

2

(α

2

T

1

x

2

+α

1

T

2

x

1

+ (α

1

−α

2

)x

1

x

2

). (5.20)

It is veriﬁed that

x

2,max

=

_

α

1

T

2

α

2

T

1

_

1/2

x

1,max

(5.21)

with

x

1,max

=

T

1

−(T

1

T

2

)

1/2

1 + (α

1

/α

2

)

1/2

, x

2,max

=

(T

1

T

2

)

1/2

−T

2

1 + (α

2

/α

1

)

1/2

. (5.22)

In particular, it is found from (5.21) and (5.22) that

T

1

= CT

1/2

1

, T

2

= CT

1/2

2

, C ≡

(α

1

T

1

)

1/2

+ (α

2

T

2

)

1/2

α

1/2

1

+α

1/2

2

. (5.23)

Introducing (5.22) in (5.17), we obtain for the maximum power output

P

max

=

α

1

α

2

(α

1/2

1

+α

1/2

2

)

2

(T

1/2

1

−T

1/2

2

)

2

. (5.24)

The eﬃciency as a function of x

1

and x

2

is

η(x, x) = 1 −

T

2

T

2

= 1 −

T

2

+x

2

T

1

−x

1

. (5.25)

Bringing it in the expressions (5.22) yields the important relation

η

max power

= 1 −

_

T

2

T

1

_

1/2

. (5.26)

It is worth to stress that, in contrast with Carnot’s result (5.5), the eﬃciency

at maximum power output relates to the square root of the ratio of temper-

atures. Both relations (5.24) and (5.26) are of high interest, especially the

second one, which is seen to depend only on the values of the temperatures

of the reservoirs, in contrast to the ﬁrst one, which is also function of the

heat transfer coeﬃcient introduced in (5.6) and (5.7).

120 5 Finite-Time Thermodynamics

As an illustration, consider a power station working, for instance, between

heat reservoirs at 565 and 25

◦

C and having an eﬃciency of 36%. We want to

evaluate this power station from the thermodynamic point of view. It follows

from (5.5) that the Carnot’s maximum eﬃciency is 64.1%; our ﬁrst opinion

on the quality of the power station would be rather negative if Carnot’s

eﬃciency is our standard for evaluation. However, according to (5.26), its

eﬃciency at maximum power is 40%. Thus, if the objective of the power

station is to work at maximum power output, it is seen that the eﬃciency

is not bad compared to the 40% eﬃciency corresponding to this situation.

We can therefore conclude that, to have realistic standards for evaluation

of an actual heat engine, one needs to go beyond Carnot’s eﬃciency and to

incorporate ﬁnite-time considerations.

We have stressed that (5.26) is independent of the heat transfer coeﬃcients,

and therefore it is rather general. However, its generality is not so universal as

Carnot’s result (5.5), because it has been derived by assuming that thermal

conduction is the only source of irreversibility, thus neglecting other possible

dissipative processes, and also because it depends on the speciﬁc form adopted

for the heat transfer law. For instance, instead of Newton’s law taken in (5.6)

and (5.7), one could assume a transfer law of the form

dQ

1

dt

= β

1

_

1

T

1

−

1

T

1

_

,

dQ

2

dt

= β

2

_

1

T

2

−

1

T

2

_

. (5.27)

In this case, it may be shown that the eﬃciency at the maximum power is

given by

η(at P

max

) =

1 + (β

1

/β

2

)

1/2

2 +

_

1 +

T

2

T

1

_

(β

1

/β

2

)

1/2

_

1 −

T

2

T

1

_

. (5.28)

Compared to (5.26), this expression depends now explicitly on the heat trans-

fer coeﬃcients. Several other heat transfer laws have been proposed in the

literature (e.g. Hoﬀmann et al. 1997).

5.1.2 Friction Losses

In the previous examples, the presence of internal dissipation has been ig-

nored. Dissipation could be, for instance, produced by friction of the piston

along the walls, heat leaks due to imperfect insulation of the reservoirs, in-

ternal losses in the working ﬂuid, such as turbulence or chemical reactions.

A simple way to introduce friction is to assume that the total force acting on

the piston is

F = pA−fv, (5.29)

where pA is the force exerted by the gas on the piston of area A and fv

represents a frictional viscous force, v being the velocity of displacement

5.1 The Finite-Time Carnot Cycle 121

of the piston and f the friction coeﬃcient. This model was investigated by

Rebhan (2002), who assumed that heat transfer eﬀects are arbitrarily fast and

that the only source of dissipation is friction. This is the situation analysed

below.

During a displacement dx, the work performed by the piston will be

dW = p dV −fv

2

(t)dt, (5.30)

and the work corresponding to one cycle is

W = Q

1

−Q

2

−f

_

c

v

2

(t)dt, (5.31)

the integration being performed over the duration of one cycle. The compu-

tation of the last term depends on the time dependence of the velocity. Let

us qualitatively write

W = Q

1

−Q

2

−

4fδD

2

τ

, (5.32)

where 2D is the total distance covered by the piston during one cycle, τ is the

duration of one cycle, and δ is a ﬁxed numerical parameter characterizing the

dependence of the velocity with respect to time (Rebhan 2002). As a result,

the power of the engine will be expressed by

P =

Q

1

−Q

2

τ

−

4fδD

2

τ

2

. (5.33)

For given values of Q

1

and Q

2

, P is a function of τ only. The maximum power

is then obtained from the condition that dP(τ)/dτ = 0, which yields for the

optimum time

τ

opt

=

8fδD

2

Q

1

−Q

2

. (5.34)

Introducing this result into (5.33), it is found that the maximum power is

P

max

=

(Q

1

−Q

2

)

2

16fδD

2

. (5.35)

The eﬃciency is given by

η =

W

Q

1

=

Q

1

−Q

2

Q

1

−

4fδD

2

τQ

1

. (5.36)

Substituting the value for the optimum time, one obtains

η(max P) =

1

2

_

1 −

Q

2

Q

1

_

=

η

Carnot

2

, (5.37)

which is of course lower than Carnot’s eﬃciency, but also lower than the

Curzon–Ahlborn’s result (5.26) derived for heat losses only.

122 5 Finite-Time Thermodynamics

Just like (5.26), expression (5.37) does not depend explicitly on the fric-

tion coeﬃcient. Combined eﬀects of heat losses and friction lead to rather

cumbersome mathematical relations, which are certainly interesting from a

practical perspective, but which do not add essential new ideas to the above

considerations.

5.2 Economical and Ecological Constraints

In practical situations, there will be a compromise between maximum eﬃ-

ciency and maximum power, and the priorities will be settled by economical

and ecological considerations. Note that, from an economic point of view, the

eﬃciency of the engine should be written as

η

economic

≡

W Pr(W)

Q

1

Pr(Q

1

)

, (5.38)

where Pr(W) is the price at which the unit of work is sold and Pr(Q

1

) is

the price at which the unit of heat which must be supplied to the engine is

bought. The economical eﬃciency will incorporate non-physical information,

as the prices of fuel and work (electrical work, for instance). As a matter of

fact, the actual economical eﬃciency is much more complicated than (5.38),

because it should also include, for instance, the investment for building the

plant and operating costs (as, for instance, the salaries of the workers, other

indirect costs as taxes, etc.). Anyway, the price of fuel plays generally a

crucial role: if fuel is expensive and the work produced by the engine may

only be sold at low price, then the situation with the highest eﬃciency, i.e.

taking most advantage of the fuel, will be the most suitable one. On the

contrary, if fuel is cheap and if the work may be sold at high price, the

situation of maximum power will be preferred from an economic point of view.

Furthermore, the higher the power output is, the faster the investments will

be recovered, whereas the higher the eﬃciency, the faster the fuel expenses

will be recovered.

Now, if ecological constraints are incorporated, the criteria may change.

For instance, the lowest is the eﬃciency, the least ecological will be the sta-

tion, because it will consume more fuel to produce the same amount of work

and therefore will produce more pollution. In short, it can be said that the

situation with the highest eﬃciency is the most ecological one, whereas, if

fuel is suﬃciently cheap, the operation at maximum power output will be

the most interesting from the economical standpoint. In general, the situa-

tion is rather complex, and will depend on particular circumstances, as for

instances, on the rate of demand in energy.

Another example where a similar conﬂict between eﬃciency and power

arises is in exothermic chemical reactions. Indeed, it is known from equilib-

rium thermodynamics that the eﬃciency of the reaction (i.e. the number of

5.2 Economical and Ecological Constraints 123

moles which are produced for a given initial quantity of reactants) is increased

when temperature is lowered, according to the Le Chatelier’s principle dis-

cussed in Chap. 1. However, from a kinetic perspective, the lower the temper-

ature is, the lower will be the reaction rate. Thus, lowering temperature has

two opposite eﬀects: it increases the eﬃciency but it diminishes the reaction

rate. Which temperature is optimal cannot be decided simply from thermo-

dynamic arguments: economical and ecological aspects must be included in

the analysis. From an economic standpoint, if the price of the reactant is low,

the most satisfactory approach is the fastest one; however, from the ecological

point of view, the faster the procedure is, the higher is the consumption of

reactant and the higher is the pollution produced by the factory.

Both power and eﬃciency must therefore be considered in the assessment

of actual heat engines. Up to now, we have discussed only two criteria, namely

power optimization and eﬃciency optimization. Other elements may be taken

into account to convey the subtleties of the interplay between economical and

ecological needs.

One proposal is to minimize the entropy production, which is related to

the loss of available work by W

lost

= T

2

∆

irr

S; the rate of entropy production

over one cycle is, after use is made of expressions (5.12) for ∆

irr

S and (5.8)

and (5.9) for t

cycle

,

σ

s

≡

∆

irr

S

t

cycle

=

Q

1

_

1

T

1

−

1

T

1

_

+Q

2

_

1

T

2

−

1

T

2

_

Q

1

α(T

1

−T

1

)

+

Q

2

α(T

2

−T

2

)

. (5.39)

In terms of x

1

and x

2

deﬁned under (5.16), this may be written as

σ

s

(x

1

, x

2

) =

α

1

α

2

T

1

T

2

x

2

1

x

2

T

2

+x

1

x

2

2

T

1

α

1

T

2

x

1

+α

2

T

1

x

2

+ (α

1

−α

2

)x

1

x

2

. (5.40)

Another ecological criterion, which is often used, is the maximization of the

quantity

Ψ = power −T

2

entropy production = P −T

2

∆

irr

S. (5.41)

By doing so, a high power (a factor of economical interest) is promoted,

together with low dissipation (a factor of ecological interest), in such a way

that one combines both economical and ecological aims. Physically, (5.41)

is related to the useful work available in a process, which is given by W =

W

R

− T

2

∆

irr

S, W

R

being the work performed by the reversible process and

T∆

irr

S the work dissipated as heat by irreversibilities. Dividing both terms

of this expression by the cycle duration t

cycle

gives back (5.41). From the

expressions (5.17) for the power output and (5.39) for the rate of entropy

production, one obtains

Ψ(x

1

, x

2

) =

α

1

α

2

T

1

T

1

(T

1

−T

2

) −(T

1

+T

2

)x

1

−2T

1

x

2

α

1

T

2

x

1

+α

2

T

1

x

2

+ (α

1

−α

2

)x

1

x

2

x

1

x

2

. (5.42)

124 5 Finite-Time Thermodynamics

The values of the variables for which this quantity becomes a maximum, in

the particular case α

1

= α

2

, are

x

1,max

=

T

1

2

T

1

−T

2

T

1

+T

2

+ 2T

1

A

, x

2,max

= Ax

1,max

, (5.43)

where

A ≡

_

T

2

(T

1

+T

2

)

2T

2

1

_

1/2

. (5.44)

Introducing (5.43) into (5.25), it is found that the corresponding eﬃciency is

η(at Ψ

max

) = η

Carnot

1 + 2

T

2

T

1

+

3

2

√

2

_

T

2

T

1

+

T

2

2

T

2

1

_

1/2

1 + 3

T

2

T

1

+ 2

√

2

_

T

2

T

1

+

T

2

2

T

2

1

_

1/2

. (5.45)

As an illustration, Table 5.1 gives the values of eﬃciency, power, entropy

production, and dissipation (product of lowest temperature times entropy

production) for a Carnot’s heat engine with T

1

= 400 K, T

2

= 300 K,

V

max

/V

min

= 2, α = 209 WK

−1

, under diﬀerent operational conditions, as

considered by Angulo-Brown (1991). Here, it has been taken into account

that Q

1

= RT

1

ln(V

max

/V

min

) and Q

2

= RT

2

ln(V

max

/V

min

).

Note the practical advantage of the Ψ optimization criterion: though it re-

duces somewhat the power availability with respect to the maximum power

(from 375 to 294 W, i.e. a 25% reduction approximately), it lowers appre-

ciably the entropy production (from 1.08 to 0.3 WK

−1

) and the dissipation

(from 325 to 93 W). Therefore, it yields a satisfactory value for the power

output, which is economically interesting, and it considerably brings down

the dissipation with respect to the maximum power situation, an appreciated

result from the ecological point of view.

Table 5.1 Values of eﬃciency, power, entropy production, and dissipation for a

Carnot’s heat engine with T

1

= 400 K, T

2

= 300 K, V

max

/V

min

= 2, α = 209 WK

−1

,

under operation conditions deﬁned by (a) maximum eﬃciency (Carnot), (b) maximum

power output, (c) minimum entropy production, and (d) maximum Ψ (from Angulo-

Brown, J. Appl. Phys. 69 (1991) 7465)

Maximum Maximum Minimum Maximum

eﬃciency power entropy production Ψ

Eﬃciency (%) 25 13 23 19.2

Power (W) 0 375 114 294

Entropy production (WK

−1

) 0 1.08 0.30 0.31

Dissipation (W) 0 325 90 93

5.3 Earth’s Atmosphere as a Non-Equilibrium System and a Heat Engine 125

5.3 Earth’s Atmosphere as a Non-Equilibrium System

and a Heat Engine

Planetary atmospheres are fascinating non-equilibrium thermodynamic sys-

tems, whose study has beneﬁted by the impressive observational data pro-

vided by space missions. It is well recognized that Earth’s atmosphere is of

utmost importance for our survival: atmosphere warming and the climate

change are among the main topics in the current agenda of Earth’s sciences

and a topic of much social, economical, and political concern since the last

decade. The atmosphere is continuously subject to energy ﬂow, because it ab-

sorbs, reﬂects, transforms, and stores shortwave solar radiation energy, and

releases back to the outer space longwave, infrared radiation energy. It repre-

sents therefore a paradigmatic example of a non-equilibrium system (Peixoto

and Oort 1984; Kleidon and Lorenz 2005). In this section, we present a sim-

pliﬁed version of the Earth energy balance, with special attention to the

understanding of global warming, and we apply ﬁnite-time thermodynam-

ics to evaluate the eﬃciency of the atmosphere as a heat engine, converting

thermal radiation energy into wind kinetic energy.

5.3.1 Earth’s Energy Balance

We present here an outline of the main factors inﬂuencing the average Earth

temperature, based on a rather simpliﬁed modelling of the global energy

balance. On the one hand, the Earth absorbs an incoming radiation power

P

abs

= (1 −r)πR

2

E

I

S

, (5.46)

where R

E

is the radius of the Earth and r is the reﬂectivity (also called

albedo), whose mean current value is 0.35. The quantity I

S

(≈1,370 Wm

−2

)

is the so-called solar constant, namely the maximum radiation ﬂux arriving

at the upper part of the atmosphere, in a region which is directly facing

the Sun, and averaged over all latitudes and seasons and over the day/night

alternation. The combination πR

2

E

is thus the transversal area of the Earth,

intercepting the solar radiation. In virtue of Wien’s law of radiation, stating

that λ

max

=2,897T

−1

µm, with T the temperature of the body emitting

radiation, the radiation arriving from the Sun, whose surface temperature is

≈6,000 K, is shortwave. Except for reﬂection and absorption in the clouds,

the atmosphere is essentially transparent to this kind of radiation, which is

directly striking the surface of the Earth.

The power emitted by the Earth is

P

emit

= 4πR

2

E

eσT

4

E

, (5.47)

where e is the emissivity, whose approximate value is 0.58, σ the Stefan–

Boltzmann constant, and T

E

the average temperature of the Earth, a quan-

tity on which we will centre our attention. The numerical factor 4 in (5.47)

126 5 Finite-Time Thermodynamics

assumes that the Earth is radiating over the whole surface, and not only the

surface facing to the Sun. According to Wien’s law, the radiation emitted by

the Earth is longwave, and behaves as infrared radiation. As commented be-

low, several gases in the atmosphere absorb and re-emit this type of radiation,

in contrast to the incoming radiation.

To ﬁnd the average temperature of the Earth, the emitted and absorbed

powers are made equal, i.e.

(1 −r)πR

2

E

I

S

= 4πR

2

E

eσT

4

E

, (5.48)

from which follows that

(1 −r)

I

S

4

= eσT

4

E

, (5.49)

and the resulting average temperature of Earth is given by

T

E

=

_

(1 −r)I

S

4eσ

_

1/4

, (5.50)

and found to be of the order of 290 K. In the balance equation (5.48), some

internal sources of heat – volcanoes and radioactive sources – have been

neglected, but should be incorporated in a more detailed analysis. According

to (5.50), the Earth reﬂectivity r and emissivity e have a deep inﬂuence on

the Earth temperature; for instance, a reduction of these parameters implies

an increase of the average temperature. If there was no greenhouse eﬀect,

whose eﬀect is to reduce e, the temperature of the Earth would be some

30

◦

C colder than it is today.

The reﬂectivity, or albedo, r and the emissivity e are fractional quantities

with values comprised between 0 and 1; for a perfect black body, one would

have r = 0 and e = 1. The albedo r and emissivity e depend on several

variables, mainly temperature and composition. The actual value of r results

from a balance between the extension of ices, snows, deserts, and clouds,

which have high values of albedo (high reﬂectivity) and the surface occupied

by forests and seas, with low values.

5.3.2 Global Warming

Global warming is nowadays the object of an enormous amount of research

at the theoretical, observational, and numerical levels. The role of human

activities and their implications on climatic changes actually deserves an

increasing international interest. We will focus here on some qualitative ideas

which allow seizing the key features of these phenomena. Truly, the ﬁrst

studies about global warming were carried out by Arrhenius in 1896, but

they did not receive a widespread attention until the end of the 1960s, when

the ﬁrst data on the increase of CO

2

in the atmosphere became available.

5.3 Earth’s Atmosphere as a Non-Equilibrium System and a Heat Engine 127

The average value of Earth’s temperature is given by (5.50). Concerning

the emissivity e, its value depends especially on the concentration of those

gases sensible to infrared radiation. The major natural infrared absorbers in

the atmosphere are water vapour, clouds, and carbon dioxide. The current

concerns are related with CO

2

and some other gases, mainly CH

4

, N

2

O, and

CFC, whose concentration has considerably increased during the last cen-

tury as a consequence of human activities. For instance, the abundance of

CO

2

in the atmosphere has increased from 315 ppmv (parts per million in

volume) in 1950 to 350 ppmv in 1990, as a consequence of fossil fuel burning

and deforestation (for the sake of comparison, let us mention that in 1750

the CO

2

atmospheric concentration was some 280 ppmv). CO

2

remains in

the atmosphere for about 500 years. The increase in CH

4

has been much less

(1% per year, at the current rate). This gas is produced by bacteria that

decompose organic matter in oxygen-poor environments, as marshes or rice

paddles, burning of forests and grasslands, or the guts of termites in the di-

gestive tracts of cattle, sheep, pigs, and other livestock. CH

4

remains in the

atmosphere for 7–10 years and each molecule is some 25 times more eﬃcient

than a molecule of carbon dioxide in warming the atmosphere. Other gases

with much eﬀect are CFC (chloroﬂuorocarbons), which also deplete ozone

in the stratosphere. The main sources are leaking air conditioners and re-

frigerators, or evaporation of some industrial solvents. CFC remain in the

atmosphere for 65–110 years, and a molecule of CFC has 10,000–20,000 times

the impact on global warming compared to a molecule of carbon dioxide.

Nitrous oxide (N

2

O) is released mainly from the breakdown of nitrogen fer-

tilizers in soil, and nitrate-contaminated groundwater. Its average stay in the

atmosphere is 150 years and the eﬀect on global warming from each molecule

is 230 times more important than a CO

2

molecule. The main contributions

to the greenhouse eﬀect are 54% CO

2

, 18% CH

4

, 14% CFC, and 6% N

2

O.

The net planetary radiative forcing is dominated by the presence of the

water vapour whose contribution to the total value of e is of the order of

60%, and whose concentration is strongly related to that of CO

2

. Note that a

relatively small variation of temperature may produce an important feedback

in the water vapour concentration. Another possible multiplicative feedback

arises from the relationship between ice extension and the albedo r: a warmer

planet will have less ice meaning a reduction in the reﬂectivity, which in

turn will produce an increase of the temperature, thus leading to a further

reduction of ice extension and so on.

The main uncertain factor inﬂuencing global warming and its impact on

climate concerns the role of clouds on the radiative balance because clouds

contribute both to warming, by absorbing infrared radiation, and to cooling,

by reﬂecting a part of the incoming radiation. The relative importance of

these two antagonistic contributions depends on the altitude of the clouds:

for low ones, warming plays the leading part, whereas for high ones, cooling is

the dominating eﬀect. Another source of query is the response of the ocean to

variations of the concentration of the greenhouse gases and to perturbations

128 5 Finite-Time Thermodynamics

in the oceanic currents distributing the heat from the Equator to the Poles.

Furthermore, since the oceans have an enormous thermal inertia, the warming

process is expected to be slow; this means that it may take many years before

warming eﬀects become perceptible.

The estimations about the CO

2

atmospheric concentration are deeply

bound to the evolution of CO

2

emissions. It is estimated that, in 2100, CO

2

concentration could be close to 550 ppmv. Solutions are not easy to ﬁnd be-

cause of the clash of interests between the necessity to reduce the use of fossil

fuels and the increasingly widespread and growing need for energy, thus lead-

ing to serious policy confrontations.

Useful information about climate change may also be drawn from a study

of the global entropy balance of Earth. The convective motions of upward hot

air and down falling cold air and several phenomena observed in our everyday

life imply the appearance of some structure and order, which must be com-

patible with a global increase of entropy of the universe. It turns out that,

globally, the Earth is a very dissipative system. Indeed, it receives photons

at T

S

≈ 6,000 K and emits them at T

E

≈ 290 K. For each incoming photon,

some 20 photons are re-emitted to the outer space. Other sources of entropy

production, as for instance heat transfer between hot and cold regions, or

turbulence of winds and oceanic currents are unavoidable and should be in-

corporated in more detailed analyses. Several models taking into account the

role of entropy production on the climate have been proposed. Based on sim-

pliﬁed atmospheric models, Nicolis and Nicolis (1980) and Paltridge (1981)

explored whether or not the climate is governed by an extremal principle re-

lated with the entropy production. They found reasonable values for the most

signiﬁcant climatic variables by maximizing the transport part of the entropy

production. We will not enter into this rather specialized topic, which throws

anyway a light on the contribution of the second law to a better understand-

ing of the global climate system. For more recent models, see for instance

Pujol and Llebot (1999) and Ozawa et al. (2003).

5.3.3 Transformation of Solar Heat into Wind Motion

In this section, we will apply ﬁnite-time thermodynamics to the Earth at-

mosphere, considered as a heat engine driven by solar radiation and pro-

ducing useful work in the form of wind motion. On the yearly average, the

quantity of heat transferred per unit time by solar radiation to the surface

of the Earth is, per unit area,

˙

Q

S

=

1 −r

4

I

S

≈ 223 Wm

−2

, (5.51)

where, as in Sect. 5.3.2, I

S

is the solar constant and r is the albedo. Since the

average energy of wind motion is approximately 7 Wm

−2

, the eﬃciency of

5.3 Earth’s Atmosphere as a Non-Equilibrium System and a Heat Engine 129

conversion of solar radiation energy into mechanical energy of wind is about

7/223 ≈ 3%.

A rough upper bound estimation for the conversion of solar heat into

work is obtained by taking for the temperature of the hot reservoir T

1

=

313 K, corresponding to the average Equator temperature of 40

◦

C, and for

the temperature of the cold reservoir T

2

= 233 K, the average temperature

at the Poles. After introducing these values in the expression of the Carnot’s

eﬃciency, one ﬁnds 25.5%, which is much higher than the 3% observed value.

Applying the maximum power eﬃciency of the Curzon–Ahlborn’s engine, the

result falls down to 14%. However, instead of Curzon–Ahlborn’s approach,

we prefer to follow an analysis by Gordon and Zarmi (1989), which is more

illustrative and pedagogical, by keeping nevertheless simplicity.

These authors investigate that the global-scale motion of wind in convec-

tive Hadley cells, namely those cells in which hot air rises towards colder

regions along the upper atmosphere, falls down on the surface of the Earth,

and moves back to hotter regions along the Earth surface. Gordon and Zarmi

model these various motions by means of a Carnot’s engine, with two isother-

mal and two adiabatic branches. The ﬁrst isothermal corresponds to the at-

mosphere absorbing solar energy at low altitudes, at temperature T

1

, not

exactly equal to the Earth surface temperature T

1

. The second isothermal

branch, at temperature T

2

, is related to the atmosphere at high altitudes,

where air rejects heat to the universe. The adiabatic branches correspond to

the rising and falling air currents. Furthermore, it is assumed that the time

spent on the isothermal branches is equal, and that the time spent on the

adiabatic branches is negligible. The values of the hotter T

1

and colder tem-

peratures T

2

of the working ﬂuid will be determined to maximize the power

conversion, in analogy with Sect. 5.1.

In the present case, the heat exchange mechanisms are of radiative type,

and therefore the analysis of Sect. 5.1, based on Newton’s law of transfer, is

not directly applicable. This justiﬁes that we re-examine the problem. The

power (per unit area) available for work will be

P =

˙

Q

1

−

˙

Q

2

=

˙

Q

S

−σ

_

T

4

1

2

+

T

4

2

2

_

, (5.52)

with the identiﬁcations

˙

Q

1

=

˙

Q

S

−

1

2

σT

4

1

,

˙

Q

2

=

1

2

σT

4

2

, and σ = 5.67

10

−8

Wm

−2

K

−4

the Stefan–Boltzmann constant, an upper dot means time

derivation. The factor

1

2

in (5.52) is a consequence of the assumption that

air spends half the time on one isotherm and half the time on the other. We

know from Sect. 5.1 that this does not exactly reﬂect the reality, but since we

are only interested in obtaining some orders of magnitude, it is not extremely

important.

On the other side, the entropy balance may be expressed as

−

˙

Q

1

T

1

+

˙

Q

2

T

2

= −

˙

Q

S

T

1

+

σT

3

1

2

+

σT

3

2

2

= 0, (5.53)

130 5 Finite-Time Thermodynamics

where it is assumed that the cycle is endoreversible and the entropy ﬂux is

simply

˙

Q

i

/T, with

˙

Q

i

given by the same relations as in (5.52).

The temperatures T

1

and T

2

are derived by maximizing the mechanical

power (5.52), subject to the constraint (5.53). Since

˙

Q

S

is ﬁxed, we may

maximize

Φ(T

1

, T

2

) ≡

σ

2

_

T

4

1

+T

4

2

_

+λ

_

˙

Q

S

T

1

−

σT

3

1

2

−

σT

3

2

2

_

, (5.54)

where λ is a Lagrange multiplier taking into account condition (5.53). The

ﬁnal results are

T

1,max

= 229 K, T

2,max

= 192 K, and P

max

= 17.1 Wm

−2

. (5.55)

The resulting eﬃciency is 7%, much closer to the actual one (3%) than the

Carnot’s and Curzon–Ahlborn’s na¨ıve estimations.

In more sophisticated formulations, the hot isotherm is considered to be

on the dayside of the Earth and the cold one on the nightside. Dissipation

of the wind energy is also introduced in some analyses. Furthermore, instead

of assuming a homogeneous model for the Earth, one may consider several

convective cells (the so-called Hadley cells) from the Equator to the Pole.

Including all these reﬁnements yields results that are very close to the actual

observed values.

5.4 Problems

5.1. Heat engine with heat leak. Assume that, parallel to the Curzon–Ahlborn’s

engine studied in Sect. 5.1, there is a direct heat leak from the hot source at

temperature T

1

and the cold source at temperature T

2

. Then the eﬃciency

must be deﬁned as

η(τ) ≡

W(τ)

Q

leak

(τ) +Q

1

(τ)

=

W(τ)

Q

leak

(τ) +

W(τ)

1−τ

,

where τ ≡ T

2

/T

1

and Q

leak

(τ) = K

leak

t

total

(τ)(T

1

−T

2

), with K

leak

the heat

transfer coeﬃcient of the leak and t

tot

(τ) the total duration of the cycle for

the Curzon–Ahlborn’s engine. Note that, in contrast with the latter, Carnot’s

eﬃciency is never reached, because due to the heat leak, Q

leak

diverges, and

the eﬃciency of this very slow cycle is very low due to the high value of the

heat leak. Find the maximum eﬃciency and the maximum power for this

engine.

5.2. The Curzon–Ahlborn’s engine. Show that, in the Curzon–Ahlborn’s en-

gine, the ratio t

1

/t

2

associated to the two isotherm branches is given by

t

1

/t

2

= (α

2

/α

1

)

1/2

.

5.4 Problems 131

5.3. Maximizing Curzon–Ahlborn’s engine. Assume that the total duration

of the cycle given by (5.8) and (5.9) is ﬁxed. Then, maximizing the power is

equivalent to maximize the diﬀerence Q

1

−Q

2

. Performing this maximization,

by introducing the constraint on the duration of the cycle by means of a

Lagrange multiplier λ, i.e. by imposing

δ

_

Q

1

−Q

2

+λ

_

Q

1

α

1

(T

1

−T

1

)

+

Q

2

α

2

(T

2

−T

2

)

__

= 0

and recalling (5.13), show that

x

1,max

x

2,max

=

_

α

2

T

1

α

1

T

2

_

1/2

,

which is the result (5.21).

5.4. Minimum entropy production criterion. Show that the minimization of

the total entropy production ∆

irr

S, under the condition of ﬁxed total duration

of the cycle, given by (5.8) and (5.9), yields

x

1,max

x

2,max

=

_

α

2

α

1

_

1/2

T

1

T

2

.

Note that this result is diﬀerent from the previous problem.

5.5. Maximization process in Curzon–Ahlborn’s engine. Assume that the en-

vironment temperature is T

0

, in such a way that the loss power is T

0

σ instead

of T

2

σ as taken in (5.41). By analogy with (5.41), study the maximization of

Ψ

0

≡ P − T

0

σ, with T

0

constant, and show that the eﬃciency at maximum

Ψ

0

is (Z. Yan, J. Appl. Phys. 73 (1993) 3583)

η(at max Ψ

0

) = 1 −

T

2

T

1

_

T

1

+T

0

T

2

+T

0

_

1/2

.

5.6. Finite-time analysis of refrigerators. One of the basic quantities in re-

frigerators is the so-called coeﬃcient of performance (COP), deﬁned as

COP ≡

heat extracted from the cold source

work supplied to the engine

.

Performing an analysis similar to that of the Curzon–Ahlborn’s heat engine,

ﬁnd the coeﬃcient of performance at maximum cooling power. Hint: see

Agrawal and Menon 1990.

5.7. Maximum power of solar thermal heat engines. A simple endoreversible

modelling of solar thermal heat engines is to assume that the heat transfer

between the hot source and the engine is given by a Stefan–Boltzmann law

132 5 Finite-Time Thermodynamics

˙

Q

1

= K(T

4

1

−T

4

1

), and that the heat transfer to the cold source is reversible.

Prove that the power–eﬃciency relation is

P(η) =

ηK

_

(1 −η)

4

T

4

1

−T

4

2

¸

(1 −η)

4

,

with η ≡ 1 − (T

2

/T

1

). Show that 4T

5

1

− 3T

2

T

4

1

− T

4

1

T

2

= 0 corresponds to

the maximum power.

5.8. Energy conversion eﬃciency of a simple photovoltaic cell. Assume that

the device is characterized by only two energy levels with a distance apart

equal to E

G

. Photons, with energy less than E

G

, simply move through the

device without leaving to it any energy. Photons with energy higher than

E

G

excite particles from the lower to the higher levels, and perform a useful

work equal to E

G

whereas the excess energy is dissipated as heat. The energy

eﬃciency may be deﬁned as

η ≡

useful power

incident power

=

E

G

_

∞

E

G

E

2

dE

exp(βE)−1

_

∞

E

G

E

3

dE

exp(βE)−1

,

where β = (k

B

T)

−1

. In the second equality, it is assumed that the energy

distribution of the incoming radiation is Planck’s black body distribution

at temperature T. (a) Show that the maximum eﬃciency is achieved for

materials whose energy gap E

G

is E

G,opt

= 2.17k

B

T. (b) Estimate E

G

if

T =5,760 K, the Sun surface temperature. (c) Evaluate the maximum eﬃ-

ciency. Truly the actual eﬃciency will be less than this value (see, for instance,

Landsberg and Tonge 1980; de Vos 1992).

5.9. Average temperature of the atmosphere. Determine the value of the av-

erage Earth temperature from (5.50) in the following cases. (a) First, assume

that the reﬂectivity r is zero and the emissivity e is unit, corresponding to the

case that the Earth behaves as a perfectly black body. (b) The reﬂectivity r

has the actual value, 35%, and e = 1 (no greenhouse eﬀect), and compare the

result with the current average temperature, namely 288 K. (c) Determine

the value of the emissivity e in order that Earth’s average temperature has

its current actual value.

5.10. Na¨ıve approach to global warming. A main feature of global warm-

ing due to increase of atmospheric CO

2

concentration is that emissivity e is

lowered, because CO

2

absorbs the infrared radiation emitted by the Earth.

Assume, for simplicity, that e = 0.58 − 0.53 ln (c

CO

2

/c

CO

2

,0

), where c

CO

2

is

the CO

2

concentration in the atmosphere and c

CO

2

,0

≈ 350 ppmv is a refer-

ence value (in parts per million in volume). Determine the temperature in-

crease if the CO

2

concentration is raised, respectively, to 700 and 1,400 ppmv.

Comment: This model is pedagogically illustrative but excessively simplistic,

5.4 Problems 133

because it ignores important feedbacks of the temperature increase on the

water vapour content, the cloud formation, and the albedo.

5.11. Sun’s and Earth’s temperatures. The current value of the solar con-

stant, namely the maximum radiation ﬂux arriving at the upper part of

the atmosphere in a region, which is directly facing the Sun, is equal to

1,367 Wm

−2

. (a) Calculate the temperature at the surface of the Sun, by

assuming that it radiates as a black body. (b) Determine the variation of

the solar constant if the temperature of the Sun is increased by 1% with

respect to its current value, obtained before. (c) To which extent will the

average temperature of the Earth change if the albedo and the emissivity

were kept constant? (Stefan–Boltzmann constant, σ = 5.67 10

−8

Wm

−2

;

distance Sun–Earth = 1.50 10

11

m, radius of the Sun = 6.96 10

8

m).

5.12. Carnot engine at ﬁnite time. A Carnot engine is working between two

heat reservoirs at 400 K and 300 K. a) Compare the maximum eﬃciency with

the eﬃciency at maximum power. b) If the engine receives 1000 J from the

hot reservoir, evaluate the entropy produced per cycle in the situation at

maximum power. c) If the duration of the cycle at maximum power is 20 s,

ﬁnd the value of the maximum power delivered by the engine. d) If the

transfer coeﬃcients between the engine and the reservoirs do not depend

on temperature, ﬁnd the eﬃciency and power of this engine when the cycle

duration is 60 seconds.

5.13. Ice comet. A spherical ice comet of 10

5

Kg describes an eccentric ellip-

tical orbit around the Sun, with major radius 2 10

8

km and minor radius

1 10

8

km. Evaluate which proportion of the mass of the comet will evaporate

during one orbit. Assume that the albedo coeﬃcient is 0.8 for ice and 0.2 for

liquid water.

5.14. Na¨ıve scaling relation for the life-span of stars. Bigger stars have

shorter lifespans than smaller stars. The physical idea is that they need

higher internal pressures, and therefore higher temperatures, to resist the

higher gravitational pressures. Since the energy emitted per unit time, which

is proportional to the loss of mass of the star per unit time in nuclear re-

actions, is proportional to T

4

, according to Stephan-Boltzmann’s law, they

will consume faster their mass. a) Make a simple estimation of the scaling

exponent α relating the lifespan of the star t

star

to the mass M of the star

through t

star

= AM

−α

, with A a constant. (Hint: equate the pressure of an

ideal gas to the average gravitational pressure determined on dimensional

grounds from G, M and R, the gravitational constant and the mass and

radius of the star. Note that if one makes the simplistic assumption that

diﬀerent stars have the same density, it is found that t

star

= AM

−7/3

). b)

According to this estimate, how long will last a star of twice the mass of the

Sun, if the total lifespan of the Sun will be of some 10

10

years? Will such a

star be able to sustain intelligent life in its planetary system, if any?

134 5 Finite-Time Thermodynamics

5.15. Minimum entropy production. a) Find the hot and cold temperature

ratios x

1

and x

2

deﬁned in (5.16) which extremize the entropy production

(5.40) (assume, for simplicity, that α

1

= α

2

). b) Determine the corresponding

value for the eﬃciency as obtained from (5.25). c) Evaluate the correspond-

ing eﬃciency and compare it with Carnot’s maximum eﬃciency and with

eﬃciency at maximum power.

Chapter 6

Instabilities and Pattern Formation

Dissipative Structures Far from Equilibrium

The notion of irreversibility is frequently associated with a tendency of the

systems towards spatial homogeneity and rest, i.e. towards equilibrium. Al-

though this observation is true in isolated systems, it is no longer veriﬁed in

closed and open systems, which may exhibit organized structures in space and

time. This organization arises as a consequence of occurrence of instabilities

in systems driven far from equilibrium by external forces. Instabilities may

be considered as phase transition phenomena, because they manifest under

the form of a discontinuous change of the physical behaviour at a critical

value of an external control parameter.

Such non-equilibrium processes are, however, rather diﬀerent from those

occurring in equilibrium phase transitions. In equilibrium thermodynamics,

there is an evident link between the second law and stability. In Chap. 1,

we have derived an explicit criterion of stability by imposing that the sec-

ond variation of the entropy is negative deﬁnite. In non-equilibrium, such a

universal criterion of stability does not exist because it is generally not pos-

sible to construct thermodynamic potentials depending on the whole set of

variables. The general problem of stability is then conventionally carried out

by submitting the reference state, for instance, a stationary non-equilibrium

state, to an arbitrary disturbance. These disturbances may be caused delib-

erately from the outside by external agents, like a temperature or pressure

gradient, or induced by irregularities or imperfections of the apparatus or

still by internal ﬂuctuations of the system (see Chap. 11).

A proof of stability is based on the time variation of the perturbation:

if the perturbation decays in a ﬁnite time, the reference state is stable, and

it is asymptotically stable if the disturbance tends to zero in the long time

limit. If the disturbance grows with time, the reference state is unstable. In

this case, the mathematical solution of the problem cannot survive in the

physical world and the behaviour of the system will be governed by another

mathematical solution: it is said that the solution suﬀers a bifurcation. When

the disturbance does neither decrease nor increase with time, one speaks

135

136 6 Instabilities and Pattern Formation

about neutral or marginal stability, and the corresponding values taken by

the relevant parameters are said to be critical.

When the perturbations remain inﬁnitesimally small, the approach is lin-

ear. Several illustrations – like B´enard–Marangoni’s convective instability,

Taylor’s instability, and Turing’s reaction–diﬀusion instability – are exam-

ined in the present chapter. One of the most interesting aspects of the sta-

bility theory is the prediction of organized structures, taking the form of

spatial and temporal patterns, which appear beyond the linear critical insta-

bility threshold. These patterns are also referred to as dissipative structures,

a term introduced by Prigogine to characterize ordered systems maintained

far from equilibrium by external constraints. The mechanisms underlying the

development of patterns are essentially non-linear and modelled by coupled

non-linear partial diﬀerential equations whose treatment is rather intricate.

Various examples are treated in the second part of this chapter.

In Sects. 6.1 and 6.2, we brieﬂy discuss the mathematical aspects of the

linear and non-linear theories of stability. When the perturbations are inﬁn-

itesimally small, the problem is linear and the solution may be decomposed

into normal modes. This technique has been made popular by the celebrated

textbook of Chandrasekhar (1961). As soon as the amplitude of the pertur-

bations becomes ﬁnite, the linear approach is no longer appropriate and must

be replaced by more sophisticated non-linear techniques or direct numerical

solutions. A detailed analysis of these methods can be found in specialized

monographs (Joseph 1976; Drazin and Reid 1981; Cross and Hohenberg 1993;

Bodenshatz et al. 2000); here we shall essentially focus on the so-called am-

plitude method which, in our opinion, is the simplest and amongst the most

powerful.

One of the most attractive aspects of the (non-linear) stability theory is the

prediction of self-organization phenomena, in the form of spatial or tempo-

ral patterns, beyond the critical threshold. Pattern formations are important

and appear in a multitude of natural phenomena being a central theme of

research in modern physics. They provide a major feature of the dynamics of

several processes observed not only in hydrodynamics but also in electricity,

optics, material science, oceanography, geophysics, astrophysics, chemistry,

and living systems. Thermal convection in ﬂuids is one of the most repre-

sentative problems and is discussed in Sect. 6.3. Indeed, most of the recent

developments about non-equilibrium processes in the context of instabilities

and chaos have been widely inspired from the study of this example.

In this chapter, the approach is more dynamical than thermodynamical,

i.e. it is based on the study of evolution equations rather than on state and

constitutive thermodynamic relations. However, from a conceptual point of

view, it is very important to understand that thermodynamics does not pre-

clude the formation of patterns provided the systems are driven far enough

from equilibrium. Such a topic represents an important contribution towards

a better knowledge of non-equilibrium systems and therefore it ﬁnds naturally

a place in this book.

6.1 The Linear Theory of Stability 137

6.1 The Linear Theory of Stability

This theory is concerned with stability of systems subject to inﬁnitesimally

small perturbations about a reference state. Most physical and chemical sys-

tems obey evolution equations of the form

da

dt

= f(a, ∇a; r, t), (6.1)

where f is a continuous and twice diﬀerentiable function depending on the

ﬁeld variables a, for instance the velocity v and the temperature T ﬁelds, and

their gradients, r designates the position vector, and t the time. Assume that

there exists a “reference” solution corresponding to some speciﬁc constraints

a = a

r

. (6.2)

Such a solution is also called unperturbed or basic, and may be time de-

pendent (for instance periodic) or time independent (i.e. stationary). The

question now arises about the stability of this particular solution. Therefore,

one examines the response of the system with respect to a disturbance a

(r, t)

so that the generalized coordinates a(r, t) become

a(r, t) = a

r

+a

(r, t), (6.3)

where the quantity a

**(r, t) is the “perturbation” which is supposed to be
**

small enough so that all non-linear terms can be neglected in (6.1). Substi-

tuting (6.3) in the evolution equations (6.1) and developing them around the

reference solution yields a linear set of diﬀerential equations of the form

da

dt

= La

, (6.4)

where the linear operator L is to be evaluated at the reference unperturbed

state. Since linear stability implies stability with respect to all possible inﬁn-

itesimal disturbances, we have to examine the reaction of the system to all

such disturbances. Practically, this will be accomplished by expressing a

as

a superposition of a complete ensemble of n normal modes, generally selected

as Fourier modes in a two-dimensional wave vector space (k = k

x

, k

y

),

a

(x, z, t) =

n

W

n

(z) exp(σ

n

t) exp(ik

n

x), (6.5)

and by examining the stability with respect to each individual mode n; x(x, y)

is the position vector in the plane normal to the z-axis. The quantity W

n

(z)

is called the amplitude and σ

n

is the growth rate of the disturbance,

σ

n

= Re σ

n

+ i Imσ

n

, (6.6)

138 6 Instabilities and Pattern Formation

generally a complex and k-dependent quantity. The k dependence of the

growth rate describes the spatial symmetry of the system; in rotationally

invariant systems, the σ

n

’s will only depend on the modulus [k[, whereas in

anisotropic systems like nematic liquid crystals there is an angle between the

direction of anisotropy and k.

The requirement that the ﬁeld equations have non-trivial solutions leads

to an eigenvalue problem for the σ

n

’s. The stability problem is completely

determined by the sign of the real part of the σ

n

’s:

• If one single Re σ

n

> 0, the system is unstable.

• If Re σ

n

< 0 for all the values of n, the system is stable.

• If Re σ

n

= 0, the stability is marginal or neutral.

In the case of marginal stability, there corresponds to each value of k a critical

value of the control parameter, say the temperature diﬀerence ∆T in B´enard’s

problem, a characteristic velocity in ﬂows through a pipe or the angular

velocity in Taylor’s problem, for which Re σ

n

= 0. All these critical values

deﬁne a curve of marginal stability, say ∆T vs. k, whose minimum (∆T

c

, k

c

)

determines the critical threshold of instability. In stability problems, it is

convenient to work with non-dimensional control parameters like the Rayleigh

number Ra in B´enard’s instability, the Reynolds number Re for the transition

from laminar to turbulent ﬂows, or the Taylor number Ta in presence of

rotation; therefore, the marginal curves and the corresponding critical values

are generally expressed in terms of these non-dimensional quantities.

In several problems, it is postulated that Re σ = 0 implies Imσ = 0, this

conjecture is called the principle of exchange of stability, which has been

demonstrated to be satisﬁed in the case of self-adjoint problems; in this case

a stationary state is attained after the onset of the instability. If Re σ = 0

but Imσ ,= 0, the onset of instability is initiated by oscillatory perturbations

and one speaks of overstability or Hopf bifurcation. This kind of instability

is observed, for instance, in rotating ﬂuids or ﬂuid layers with a deformable

interface. The condition Re σ

n

> 0 for at least one value of n is a suﬃcient

condition of instability; on the contrary, even when all the eigenvalues are

such that Re σ

n

< 0, one cannot conclude in favour of stability as one can-

not exclude the possibility that the system is unstable with respect to ﬁnite

amplitude disturbances. It is therefore worth to stress that a linear stability

analysis predicts only suﬃcient conditions of instability.

6.2 Non-Linear Approaches

As soon as the amplitude of the disturbance is ﬁnite, the linear approach is

not appropriate and must be replaced by non-linear theories. Among them

one may distinguish the “local” and the “global” ones. In the latter, the de-

tails of the motion and the geometry of the ﬂow are omitted, instead attention

6.2 Non-Linear Approaches 139

is focused on the behaviour of global quantities, generally chosen as a positive

deﬁnite functional. A typical example is Lyapounov’s function; according to

Lyapounov’s theory, the system is stable if there exists a functional Z satisfy-

ing Z > 0 and dZ/dt ≤ 0. In classical mechanics, an example of Lyapounov’s

function is the Hamiltonian of conservative systems. Glansdorﬀ and Prigogine

(1971) showed that the second variation of entropy δ

2

S provides an exam-

ple of Lyapounov’s functional in non-equilibrium thermodynamics. The main

problems with Lyapounov’s theory are:

1. The diﬃculty to assign a physical meaning to the Lyapounov’s functional

2. The fact that a given situation can be described by diﬀerent functionals

3. That in practice, it yields only suﬃcient conditions of stability

We do no longer discuss this approach and invite the interested reader to

consult specialized works (e.g. Movchan 1959; Pritchard 1968; Glansdorﬀ

and Prigogine 1971). Here we prefer to concentrate on the more standard

“local” methods where it is assumed that the perturbation acts at any point

in space and at each instant of time. We have seen that the solution of the

linearized problem takes the form exp(σ

n

t) and that instability occurs when

the growth rate becomes positive, or equivalently stated, when the dimension-

less control parameter R exceeds its critical value R

c

. For values of R > R

c

,

the hypothesis of small amplitudes is no longer valid as non-linear terms be-

come important and will modify the exponential growth of the disturbances.

Another reason for taking non-linear terms into account is that the linear ap-

proach predicts that a whole spectrum of horizontal wave numbers become

unstable. This is in contradiction with experimental observations, which show

a tendency towards simple cellular patterns indicating that only one single

wave number, or a small band of wave numbers, is unstable.

Non-linear methods are therefore justiﬁed to interpret the mechanisms oc-

curring above the critical threshold. The problem that is set up is a non-linear

eigenvalue problem. Unfortunately, no general method for solving non-linear

diﬀerential equations in closed form has been presented and this has moti-

vated the development of perturbation techniques. A widely used approach

is the so-called amplitude method initiated by Landau (1965) and developed

by Segel (1966), Stuart (1958), Swift and Hohenberg (1977), and many oth-

ers. It is essentially assumed that the non-linear disturbances have the same

form as the solution of the linear problem with an unknown time-dependent

amplitude. Explicitly, the solutions will be expressed in terms of the eigen-

vectors W(z) of the linear problem in the form

a

(x, z, t) = A(t) exp(ik x)W(z), (6.7)

where A(t) denotes an unknown amplitude, generally a complex quantity.

In the linear approximation, A(t) is proportional to exp(σt) and obeys the

linear diﬀerential equation

dA(t)

dt

= σA(t), (6.8)

140 6 Instabilities and Pattern Formation

whereas in the non-linear regime, projection of (6.8) on the space of the W’s

leads to a coupled system of non-linear ordinary diﬀerential equations for the

amplitudes

dA(t)

dt

= σA(t) +N(A, A), (6.9)

where N(A, A) designates the non-linear contributions. Practically, the equa-

tions are truncated at the second or third order. A simple example is provided

by the following Landau relation (Drazin and Reid 1981)

dA

dt

= σA−l

c

[A[

2

A, (6.10)

where l

c

is a complex constant depending on the system to be studied and

[A[

2

= AA

∗

with A

∗

the complex conjugate of A. In (6.10), one has imposed

the constraint A = −A reﬂecting the inversion symmetry of the ﬁeld variables

like the velocity and temperature ﬁelds. This invariance property is destroyed

and additional quadratic terms in A

2

will be present when some material

parameters like viscosity or surface tension are temperature dependent. To

take into account some spatial eﬀects like the presence of lateral boundaries,

it may be necessary to complete the above relation (6.9) by spatial terms in

A or independent terms.

By multiplying (6.10) by A

∗

and adding the complex conjugate equation,

one arrives at

d [A[

dt

2

= 2(Re σ) [A[

2

−2l [A[

4

, (6.11)

where l is the real part of l

c

. If A

0

designates the initial value, the solution

of (6.11) is

[A[

2

=

A

2

0

l

Re σ

A

2

0

+

_

1 −

l

Re σ

A

2

0

_

exp[−2(Re σ)t]

. (6.12)

(1) Let us ﬁrst examine what happens for l > 0. When (Re σ) < 0, the system

relaxes towards the reference state A = 0 which is therefore stable; in

contrast, for (Re σ) > 0, the solution (6.12) tends, for t → ∞, to a

stationary solution [A

s

[ given by

[A

s

[ = (Re σ/l)

1/2

, (6.13)

which is independent of the initial value A

0

. This is a supercritical sta-

bility, the reference ﬂow becomes linearly unstable at the critical point

Re σ = 0, or equivalently at R = R

c

, and bifurcates on a new steady

stable branch with an amplitude tending to A

s

. When the bifurcation is

supercritical, the transition between the successive solutions is continu-

ous and is called a pitchfork bifurcation as exhibited by Fig. 6.1.

It is instructive to develop Re σ around the critical point in terms of the

wave number k and the dimensionless characteristic number R so that

6.2 Non-Linear Approaches 141

Fig. 6.1 Supercritical pitchfork bifurcation: the solution A = 0 is linearly stable for

R < R

c

but linearly unstable for R > R

c

, the branching of the curve at the critical

point R = R

c

is called a bifurcation. Unstable states are represented by dashed lines

and stable states are represented by solid lines

Re σ = α(R −R

c

) +β(k −k

c

) (k −k

c

) + , (6.14)

where α is some positive constant. When R < R

c

, all perturbations

are stable with Re σ < 0; at R = R

c

, the system is marginally stable

and when R increases above R

c

, the system becomes linearly unstable.

Combining (6.14) with (6.13) results in

A

s

∼ (R −R

c

)

1/2

as R →R

c

, (6.15)

indicating that the amplitude A

s

of the steady solution is proportional to

the square root of the distance from the critical point. There is a strong

analogy with a phase transition of second order where the amplitude

A

c

of the critical mode plays the role of the order parameter and the

exponent 1/2 in (6.15) is the critical exponent.

(2) Let us now examine the case l < 0. If Re σ > 0, both terms of Lan-

dau’s equation (6.11) are positive and [A[ grows exponentially; it follows

from (6.12) that [A[ is inﬁnite after a ﬁnite time t = (2Re σ)

−1

ln[1 −

(Re σ)/(lA

2

0

)], however this situation never occurs in practice because in

this case it is necessary to include higher-order terms in [A[

6

, [A[

8

, . . .

in Landau’s equation and generally no truncation is allowed. A more

realistic situation corresponds to Re σ < 0; now, the two terms in the

right-hand side of (6.12) are of opposite sign. Depending on whether

A

0

is smaller or larger than [A

s

[ given by (6.12), we distinguish two

diﬀerent behaviours; for A

0

< [A

s

[, the solution given in (6.12) shows

that [A[ ≈ exp[(2Re σ)t] and tends to zero as t → ∞; in contrast

for A

0

> [A

s

[, the denominator of (6.12) becomes inﬁnite after a time

t = (2Re σ)

−1

ln[1 − Re σ/lA

2

0

] and [A[ → ∞ (see Fig. 6.2). In this case,

142 6 Instabilities and Pattern Formation

Fig. 6.2 Time dependence of the amplitude for two diﬀerent initial values A

0

in

the case of a subcritical instability

the reference state is stable with respect to inﬁnitesimally small distur-

bances but unstable for perturbations with amplitude greater than the

critical value A

s

, which appears as a threshold value. This situation is

referred to as subcritical or metastable, by using the vocabulary of the

physicists.

In some systems, as for instance non-Boussinesq ﬂuids, where the transport

coeﬃcients like the viscosity or the surface tension are temperature depen-

dent, the symmetry A = −A is destroyed and the amplitude equation takes

the form

dA

dt

= σA+gA

2

−lA

3

, (6.16)

when A is assumed to be real, g and l are positive constants characterizing

the system. This form admits three steady solutions A

s

= 0 and A

1,2

given by

A

1,2

=

g ±

_

g

2

+ 4σl

2l

, (6.17)

and they are represented in Fig. 6.3 wherein the amplitude A

s

is sketched as

a function of the dimensionless number R.

For R < R

G

, the basic ﬂow is globally stable which means that all pertur-

bations, even large, decay ultimately; for R

G

< R < R

c

, the system admits

two stable steady solutions A

s

= 0 and the branch GD whereas CG is un-

stable. At R = R

c

, the system becomes unstable for small perturbations and

we are faced with two possibilities: either there is a continuous transition

towards the branch CF which is called a transcritical bifurcation, character-

ized by the intersection of two bifurcation curves, or there is an abrupt jump

to the stable curve DE, the basic solution “snaps” through the bifurcation

to some ﬂow with a larger amplitude. By still increasing R, the amplitude

will continue to grow until a new bifurcation point is met. If, instead, the

6.3 Thermal Convection 143

Fig. 6.3 Subcritical instability: the system is stable for inﬁnitesimally small per-

turbations but unstable for perturbations with amplitude larger than some critical

value. Solid and dot lines refer to stable and unstable solutions, respectively

amplitude is gradually decreased, one moves back along the branch EDG up

to the point G where the system falls down on the basic state A = 0 iden-

tiﬁed by the point H. The cycle CDGH is called a hysteresis process and is

reminiscent of phase transitions of the ﬁrst order.

In the present survey, it was assumed that the amplitude equation was

truncated at order 3. In presence of strong non-linearity, i.e. far from the

linear threshold, such an approximation is no longer justiﬁed and the intro-

duction of higher-order terms is necessary, however this would result in rather

intricate and lengthy calculations. This is the reason why model equations,

like the Swift–Hohenberg equations (1977) or generalizations of them (Cross

and Hohenberg 1993; Bodenshatz et al. 2000), have been recently proposed.

Although such model equations cannot be derived directly from the usual

balance equations of mass, momentum, and energy, they capture most of the

essential of the physical behaviour and have become the subject of very in-

tense investigations. Recent improvements in the performances of numerical

analysis have fostered the resolution of stability problems by direct integra-

tion of the governing equations. Although such approaches are rather heavy,

costly, and mask some interesting physical features, they are useful as they

may be regarded as careful numerical control of the semi-analytical methods

and associated models.

6.3 Thermal Convection

Fluid motion driven by thermal gradients, also called thermal convection, is

a familiar and important process in nature. It is far from being an academic

subject. Beyond its numerous technological applications, it is the basis for

the interpretation of several phenomena as the drift of the continental plates,

144 6 Instabilities and Pattern Formation

the Sun activity, the large-scale circulations observed in the oceans, the at-

mosphere, etc. As a prototype of thermal convection, we shall examine the

behaviour of a thin ﬂuid layer enclosed between two horizontal surfaces whose

lateral dimensions are much larger than the width of the layer. The two hori-

zontal bounding planes are either rigid plates or stress-free surfaces, the lower

surface is uniformly heated so that the ﬂuid is subject to a vertical tempera-

ture gradient. If the temperature gradient is suﬃciently small, heat is trans-

ferred by conduction alone and no motion is observed. When the temperature

diﬀerence between the two plates exceeds some critical value, the conduction

state becomes unstable and motion sets in. The most inﬂuential experimental

investigation on thermal convection dates back to B´enard (1900). The ﬂuid

used by B´enard was molten spermaceti, a whale’s non-volatile viscous oil, and

the motion was made visible by graphite or aluminium powder. In B´enard’s

original experiment, the lower surface was a rigid plate but the upper one

was open to air, which introduces an asymmetry in the boundary conditions

besides surface tension eﬀects. The essential result of B´enard’s experiment

was the occurrence of a stable, regular pattern of hexagonal convection cells.

Further investigations showed that the ﬂow was ascending in the centres of

the cells and descending along the vertical walls of the hexagons. Moreover,

optical investigations revealed that the ﬂuid surface was slightly depressed at

the centre of the cells.

A ﬁrst theoretical interpretation of thermal convection was provided by

Rayleigh (1920), whose analysis was inspired by the experimental observa-

tions of B´enard. Rayleigh assumed that the ﬂuid was conﬁned between two

free perfectly heat conducting surfaces, and that the ﬂuid properties were

constant except for the mass density. In Rayleigh’s view, buoyancy is the

single responsible for the onset of instability. By assuming small inﬁnitesimal

disturbances, he was able to derive the critical temperature gradient for the

onset of convection together with the wave number for the marginal mode.

However, it is presently recognized that Rayleigh’s theory is not adequate to

explain the convective mechanism investigated by B´enard. Indeed in B´enard’s

set up, the upper surface is in contact with air, and surface tractions originat-

ing from surface tension gradients may have a determinant inﬂuence on the

onset of the ﬂow. By using stress-free boundary conditions, Rayleigh com-

pletely disregarded this eﬀect. It should also be realized that surface tension

is not a constant but that it may depend on the temperature or (and) the

presence of surface contaminants. This dependence is called the capillary or

the Marangoni eﬀect after the name of the nineteenth-century Italian investi-

gator. The importance of this eﬀect was only established more than 40 years

later after Rayleigh’s paper by Block (1956) from the experimental point of

view. Pearson (1958) made the ﬁrst theoretical study about the inﬂuence of

the variation of surface tension with temperature on thermal convection. The

predominance of the Marangoni eﬀect in B´enard’s original experiment is now

admitted beyond doubt and conﬁrmed by experiments conducted recently in

space-ﬂight missions where gravity is negligible. When only buoyancy eﬀects

6.3 Thermal Convection 145

are accounted for, the problem is generally referred to as Rayleigh–B´enard’s

instability while B´enard–Marangoni is the name used to designate surface

tension-driven instability. When both buoyancy and surface tension eﬀects

are present, one speaks about the Rayleigh–B´enard–Marangoni’s instability.

6.3.1 The Rayleigh–B´enard’s Instability:

A Linear Theory

We are going to study the instabilities occurring in a viscous ﬂuid layer of

thickness d (between a few millimetres and a few centimetres) and inﬁnite

horizontal extent limited by two horizontal non-deformable free surfaces, the

z-axis is pointing in the opposite direction of the gravity acceleration g. The

ﬂuid is heated from below with T

h

and T

c

, the temperatures of the lower and

upper surfaces, respectively (see Fig. 6.4). The mass density ρ is assumed to

decrease linearly with the temperature according to the law

ρ = ρ

0

[1 −α(T −T

0

)], (6.18)

where T

0

is an arbitrary reference temperature, say the temperature of the

laboratory, and α the coeﬃcient of thermal expansion, generally a positive

quantity except for water around 4

◦

C. For ordinary liquids, α is of the order

of 10

−3

–10

−4

K

−1

.

When the temperature diﬀerence ∆T = T

h

−T

c

(typically not more than

a few

◦

C) between the two bounding surfaces is lower than some critical

value, no motion is observed and heat propagates only by conduction inside

the ﬂuid. However by further increasing ∆T, the basic heat conductive state

becomes unstable at a critical value (∆T)

crit

and matter begins to perform

bulk motions which, in rectangular containers, take the form of regular rolls

aligned parallel to the short side as visualized in Fig. 6.5, this structure is

referred to as a roll pattern. Note that the direction of rotation of the cells is

unpredictable and uncontrollable, and that two adjacent rolls are rotating in

opposite directions.

Fig. 6.4 Horizontal ﬂuid layer submitted to a temperature gradient opposed to the

acceleration of gravity g

146 6 Instabilities and Pattern Formation

Fig. 6.5 Convective rolls in Rayleigh–B´enard’s instability

A qualitative interpretation of the onset of motion is the following. By

submitting the ﬂuid layer to a temperature diﬀerence, one generates a tem-

perature and a density gradient. A ﬂuid droplet close to the hot lower plate

has a lower density than everywhere in the layer, as density is generally a

decreasing function of temperature. As long as it remains in place, the ﬂuid

parcel is surrounded by particles of the same density, and all the forces acting

on it are balanced. Assume now that, due to a local ﬂuctuation, the droplet is

slightly displaced upward. Being surrounded by cooler and denser ﬂuid, it will

experience a net upward Archimede’s buoyant force proportional to its vol-

ume and the temperature diﬀerence whose eﬀect is to amplify the ascending

motion. Similarly a small droplet initially close to the upper cold plate and

moving downward will enter a region of lower density and becomes heavier

than the surrounding particles. It will therefore continue to sink, amplifying

the initial descent. What is observed in the experiments is thus the result

of these upward and downward motions.

However, experience tells us that convection does not appear whatever

the temperature gradient as could be inferred from the above argument. The

reason is that stabilizing eﬀects oppose the destabilizing role of the buoyancy

force; one of them is viscosity, which generates a friction force directed oppo-

site to the motion, the second one is heat diﬀusion, which tends to spread out

the heat contained in the droplet towards its environment reducing the tem-

perature diﬀerence between the droplet and its surroundings. This explains

why a critical temperature diﬀerence is necessary to generate a convective

ﬂow: motion will start as soon as buoyancy overcomes the dissipative eﬀects

of viscous friction and heat diﬀusion. These eﬀects are best quantiﬁed by the

introduction of the thermal diﬀusion time and the viscous relaxation time

τ

χ

= d

2

/χ, τ

ν

= d

2

/ν, (6.19)

where χ is the thermal diﬀusivity, ν the kinematic viscosity, and d a scaling

length, τ

χ

is the time required by the ﬂuid to reach thermal equilibrium

with its environment, τ

ν

is related to the time needed to obtain mechanical

6.3 Thermal Convection 147

equilibrium. Another relevant timescale is the buoyant time, i.e. the time that

a droplet, diﬀering from its environment by a density defect δρ = ρ

0

α∆T,

needs to travel across a layer of thickness d,

τ

2

B

= d/(αg∆T). (6.20)

This result is readily derived from Newton’s law of motion ρ

0

d

2

z/dt

2

= gδρ

for a small volume element; a large value of ∆T means that the buoyant

time is short. To give an order of magnitude of these various timescales,

let us consider a shallow layer of silicone oil characterized by d = 10

−3

m,

ν = 10

−4

m

2

s

−1

, χ = 10

−7

m

2

s

−1

, it is then found that τ

ν

= 10

−2

s and

τ

χ

= 10 s.

The relative importance of the buoyant and dissipative forces is obtained

by considering the ratios τ

ν

/τ

B

and τ

χ

/τ

B

or, since they occur simultaneously,

through the so-called dimensionless Rayleigh number,

Ra =

τ

ν

τ

χ

τ

2

B

=

αg∆Td

3

νχ

. (6.21)

The Rayleigh number can therefore be viewed as the ratio between the desta-

bilizing buoyancy force and the stabilizing eﬀects expressed by the viscous

drag and the thermal diﬀusion; convection will start when Rayleigh number

exceeds some critical value (Ra)

c

. For Ra < (Ra)

c

, the ﬂuid remains at rest

and heat is only transferred by conduction, for Ra > (Ra)

c

, there is a sudden

transition to a complex behaviour characterized by the emergence of order in

the system. The ratio between the dissipative processes is measured by the

dimensionless Prandtl number deﬁned as

Pr = τ

χ

/τ

ν

= ν/χ, (6.22)

for gases Pr ∼ 1, for water Pr = 7, for silicone oils Pr is of the order of 10

3

,

and for the Earth’s mantle Pr ∼ 10

23

.

In a linear stability approach, the main problem is the determination of

the marginal stability curve, i.e. the curve of Ra vs. the wave number k at

σ = 0. The one corresponding to Rayleigh–B´enard’s instability is derived in

the Box 6.1.

Box 6.1 Marginal Stability Curve

The mathematical analysis is based on the equations of ﬂuid mechanics writ-

ten within the Boussinesq approximation. This means ﬁrst that the density

is considered to be constant except in the buoyancy term; second that all the

material properties as viscosity, thermal diﬀusivity, and thermal expansion

coeﬃcient are temperature independent; and third that mechanical dissi-

pated energy is negligible. The governing equations of mass, momentum,

and energy balance are then given by

∇ v = 0,

148 6 Instabilities and Pattern Formation

_

∂v

∂t

+v ∇v

_

= −

1

ρ

0

∇p + [1 −α(T −T

0

)]g +ν∇

2

v, (6.1.1)

∂T

∂t

+v ∇T = χ∇

2

T,

where use has been made of the equation of state (6.18) and where

v(v

x

, v

y

, v

z

) and p designate the velocity and pressure ﬁelds, respectively.

The set (6.1.1) represents ﬁve scalar partial diﬀerential equations for the

ﬁve unknowns p, T and v

x

, v

y

, v

z

.

In the basic unperturbed state, the ﬂuid is at rest and temperature is

conveyed by conduction, so that the solutions of (6.1.1) are simply

v

r

= 0, T

r

= T

h

−

∆T

d

z,

∂p

r

∂z

= −ρ

0

g[1 −α(T

r

−T

0

)], (6.1.2)

where subscript r refers to the unperturbed reference solution, z denotes the

vertical coordinate measured positive upwards with z = 0 corresponding to

the lower boundary and x, y the horizontal coordinates, and ∆T = T

h

−T

c

is the positive temperature diﬀerence between the lower and upper bound-

aries. Designating by v

= v −0, T

= T −T

r

, p

= p−p

r

the inﬁnitesimally

small perturbations of the basic state, we can linearize (6.1.1) and obtain

the following set for the perturbed ﬁelds

∇ v

= 0, (6.1.3)

∂v

∂t

= −

1

ρ

0

∇p

+αT

ge

z

+ν∇

2

v

, (6.1.4)

∂T

∂t

+v

∇T

r

= χ∇

2

T

, (6.1.5)

where e

z

is the unit vector pointing opposite to g. We now determine the

corresponding boundary conditions. If we assume that the thermal conduc-

tivity at the limiting surfaces is much higher than in the ﬂuid itself, any

thermal disturbance advected by the ﬂuid will be instantaneously smoothed

out so that T

**will vanish at the bounding surfaces. Since the horizontal
**

boundaries are assumed to be free surfaces, the shearing stress is zero at

the surface; when use is made of the equation of continuity (6.1.3), this

condition is identical to setting ∂

2

v

z

/∂z

2

= 0 together with v

z

= 0, as the

surfaces are non-deformable. Summarizing, the boundary conditions are

v

z

= ∂

2

v

z

/∂z

2

= 0, T

**= 0 at perfectly heat conducting free surfaces.
**

(6.1.6)

In the case of rigid boundaries, the no-slip condition imposes that all the

components of the velocity are zero v

x

= v

y

= v

z

= 0, which combined with

the continuity condition yields

v

z

= ∂v

z

/∂z = 0, T

**= 0 at perfectly heat conducting rigid walls.
**

(6.1.7)

6.3 Thermal Convection 149

At adiabatically isolated walls, the condition T

= 0 will be replaced by

∂T

**/∂z = 0, expressing that the surface is impermeable to heat ﬂow. When
**

the boundary surface is neither perfectly heat conducting nor adiabatically

isolated, heat transfer is governed by Newton’s cooling law −λ∂T/∂z =

h(T −T

∞

) where λ is the heat conductivity of the ﬂuid, h the so-called heat

transfer coeﬃcient, and T

∞

the temperature of the outside world.

A further simpliﬁcation of the set (6.1.3)–(6.1.5) is obtained by applying

twice the operator rot (≡ ∇) on the momentum equation and using the

continuity equation, we are then left with the two following relations in the

two unknowns v

z

and T

∂

∂t

(∇

2

v

z

) = αg∇

2

1

T

+ν∇

4

v

z

, (6.1.8)

∂T

∂t

= βv

z

+χ∇

2

T

, (6.1.9)

where ∇

2

1

≡ ∂

2

/∂x

2

+ ∂

2

/∂y

2

denotes the horizontal Laplacian whereas

β = ∆T/d. We now make the variables dimensionless by introducing the

following scaling

X =

x

d

, Y =

y

d

, Z =

z

d

,

ˆ

t =

ν

d

2

t, w =

d

ν

v

z

, θ =

χ

ν

T

∆T

, (6.1.10)

and solve (6.1.8) and (6.1.9) with a normal mode solution of the form

w = W(Z) exp[i(k

x

X +k

y

Y )] exp(σ

ˆ

t), (6.1.11)

θ = Θ(Z) exp[i(k

x

X +k

y

Y )] exp(σ

ˆ

t), (6.1.12)

where W(Z) and Θ(Z) are the amplitudes of the perturbations, k

x

and k

y

are the dimensionless wave numbers in the directions x and y, respectively,

and σ is the dimensionless growth rate. Since the present problem is self-

adjoint, it can be proved (Chandrasekhar 1961) that σ is a real quantity.

There exist several ways to make the variables dimensionless, the choice

made here seems to be one of the most preferred. Sometimes, it is also

preferable to describe the horizontal periodicity of the ﬂow by means of the

wavelength λ = 2π/k rather than the wave number as it provides a direct

measure of the dimensions of the cells.

Substitution of solutions (6.1.11) and (6.1.12) in (6.1.8) and (6.1.9)

leads to the following amplitude diﬀerential equations

(D

2

−k

2

)(D

2

−k

2

−σ)W = Ra k

2

Θ, (6.1.13)

(D

2

−k

2

−σ Pr)Θ = −W, (6.1.14)

where D stands for d/dZ and k

2

= k

2

x

+k

2

y

, Ra and Pr denote the dimen-

sionless Rayleigh and Prandtl numbers deﬁned by (6.21) and (6.22). The

boundary conditions corresponding to two perfectly heat conducting free

surfaces are

150 6 Instabilities and Pattern Formation

W = D

2

W = 0, Θ = 0 at Z = 0 and Z = 1. (6.1.15)

The latter are satisﬁed for solutions of the form W = Asin πZ, Θ =

Bsin πZ, which substituted in (6.1.13) and (6.1.14) lead to the following

algebraic equations for the two unknowns A and B:

A+ (π

2

+k

2

+σ)B = 0, (6.1.16)

(π

2

+k

2

)(π

2

+k

2

+σPr

−1

)A−k

2

Ra B = 0.

Non-trivial solutions demand that the determinant of the coeﬃcients

vanishes, which results in the following dispersion relation between k, σ,

Ra, and Pr:

(π

2

+k

2

)σ

2

+ (π

2

+k

2

)

2

(1 +Pr)σ +Pr[(π

2

+k

2

)

3

−k

2

Ra] = 0. (6.1.17)

By setting σ = 0 in (6.1.17), one obtains the marginal curve (Ra)

0

vs. k

determining the Rayleigh number at the onset of convection (Fig. 6.6); it is

independent of the Prandtl number and given by

(Ra)

0

=

(π

2

+k

2

)

3

k

2

. (6.23)

Fig. 6.6 Marginal stability curve for Rayleigh–B´enard’s instability in a horizontal

ﬂuid layer limited by two stress-free perfectly heat conducting surfaces

6.3 Thermal Convection 151

The solutions of (6.1.17) can be written as

σ =

(π

2

+k

2

)(1 +Pr)

2

_

−1 ±

_

1 +

4(Ra −(Ra)

0

)

(Ra)

0

(1 +Pr)(1 +Pr

−1

)

_

1/2

_

,

(6.24)

which shows explicitly that the growth rate is a function of k, Ra, and Pr.

For Ra < (Ra)

0

, one has σ < 0 (stability) while for Ra > (Ra)

0

, σ > 0

(instability). It follows also from (6.24) that σ is real when Ra > 0 and

that σ < 0 when Ra < 0; this latter result indicates that the system is

unconditionally stable by heating from above.

For slightly supercritical conditions, where Ra is close to (Ra)

0

, (6.24)

reads as

σ = (π

2

+k

2

)

Ra −(Ra)

0

(Ra)

0

(1 +Pr

−1

)

, (6.25)

which shows that, for Ra larger than (Ra)

0

, the amplitude of the disturbances

ampliﬁes exponentially and the basic state is unstable. However, for such

values, the non-linear terms become important and the linear analysis ceases

to be valid.

The minimum value of the marginal curve is obtained from relation (6.23)

by diﬀerentiation with respect to k; setting this result equal to zero gives the

critical wave number k

c

at which the curve (Ra)

0

(k) is minimum, and the

corresponding critical Rayleigh number (Ra)

c

. For the present problem, it is

found that

k

c

= π/

√

2 = 2.21, (Ra)

c

= 27π

4

/4 = 657.5 (free–free boundary conditions).

(6.26)

The critical Rayleigh number allows us to determine the critical temperature

diﬀerence at which the system changes from the state of rest to the state

of cellular motion, the critical wave number provides information about the

horizontal periodicity of the patterns at the onset of convection, k

c

represents

the most dangerous mode picked up by the ﬂuid. For other boundary condi-

tions, the calculations are more complicated but the procedure remains valid,

the marginal curves will have approximately the same form as in Fig. 6.6 with

critical values given by

k

c

= 2.68, (Ra)

c

= 1, 100.6 (rigid–free boundaries),

k

c

= 3.117, (Ra)

c

= 1, 707.7 (rigid–rigid boundaries).

As expected, stability is reinforced (larger (Ra)

c

value) in presence of rigid

surfaces as the ﬂuid motion is more strongly inhibited by the viscous forces.

At the same time the dimensions of the cells (larger k

c

value) are diminished:

more energy must now be dissipated to compensate for the larger release of

energy by buoyancy, clearly narrower cells are associated with greater dis-

sipation and energy release. It is also worth to note that, for Ra > (Ra)

c

152 6 Instabilities and Pattern Formation

and according to the linear theory, a continuous spectrum of modes becomes

unstable from which follows that the observed pattern should be very intri-

cate. This is in contradiction with the observation that the ﬂow prefers rather

simple cellular forms; the reason for this discrepancy must once more be at-

tributed to the omission of non-linear terms. Moreover, the linear theory is

unable to predict the particular pattern (either rolls, squares, or hexagons)

selected by the ﬂuid; this is so because the eigenvalue problem is degenerate,

which means that to one eigenvalue Ra there corresponds an inﬁnite number

of possible patterns with the same wave number k. The reason why a partic-

ular pattern is selected can only be understood from a non-linear approach,

which shows that, for the present problem, two-dimensional parallel rolls are

the preferred patterns as conﬁrmed by experimental observations.

6.3.2 The Rayleigh–B´enard’s Instability:

A Non-Linear Theory

We now examine the behaviour of the amplitude of the disturbance beyond

the critical point. Returning to the linear theory where the amplitude is

supposed to behave as A(t) ∼ exp σt, we may write that, at threshold, the

relevant diﬀerential equation is

dA

dt

= σA, (6.27)

where σ is given by (6.25). For supercritical Rayleigh numbers, the amplitude

will then increase exponentially but non-linear self-interaction between modes

becomes important giving raise to higher-order terms in A

n

. We are then led

to an amplitude equation of the form suggested by Landau, i.e.

dA

dt

= σA−lA

3

, (6.28)

where l is a positive constant to be determined from the boundary conditions

whereas σ is the growth rate corresponding to the most dangerous mode k

c

(= π/

√

2 for free–free boundaries). In virtue of (6.25) and for large values of

Pr as in silicone oils, σ is given by

σ = 3k

2

c

ε, (6.29)

in which

ε = [Ra −(Ra)

c

]/(Ra)

c

(6.30)

measures the relative distance from the critical point. There is no term in A

2

in (6.28) because it is assumed that the convective pattern is such that, by

reversing the ﬂuid velocities, the same pattern is observed; this implies that

(6.28) must be invariant with respect to the symmetry A = −A. Of course

6.3 Thermal Convection 153

this is only true for two-dimensional rolls and it is not so for three-dimensional

patterns as hexagonal cells. Solving (6.28) for steady conditions, one obtains

A

s

= (σ/l)

1/2

= k

c

(3ε/l)

1/2

, (6.31)

from which it follows that the steady amplitude A

s

is proportional to ε

1/2

; this

result is typical of a supercritical bifurcation as represented on Fig. 6.1. To

examine the stability of the steady solution, let us superpose an inﬁnitesimal

disturbance A

such that A = A

s

+A

**. Substituting this expression in (6.28)
**

and using dA

s

/dt = 0 leads to

dA

dt

= −2σA

= −6k

2

c

εA

, (6.32)

and, after integration,

A

∼ exp[−6k

2

c

εt]. (6.33)

The perturbation decays for ε > 0, i.e. for Ra > (Ra)

c

, and it is concluded

that two-dimensional rolls are stable for perturbations with a wave number

equal to the critical one. When several modes are taken into account as in

the B´enard–Marangoni’s problem which is treated in Sect. 6.3.3, the problem

is much more complicated because various modes will interact and instead of

(6.28), we are led to a coupled set of non-linear equations of the form

dA

n

dt

= σA

n

−A

n

m

l

mn

A

2

m

−lA

3

n

(m, n = 1, 2, . . .). (6.34)

An example of such kind of equations is the famous Lorenz model (1963),

which is widely used in meteorology and which is at the essence of the theory

of chaos. The explicit form of Lorenz set of equations is

dA

1

dt

= Pr(A

2

−A

1

),

dA

2

dt

= rA

1

−A

2

−A

1

A

3

,

dA

3

dt

= −bA

3

+A

1

A

2

,

(6.35)

where r = Ra/(Ra)

c

and b = 4/(1 + 4k

2

). The coeﬃcient A

1

describes the

velocity of the ﬂuid particles, A

2

the temperature ﬂuctuations, and A

3

a

horizontally averaged temperature mode. A detailed discussion of the Lorenz

model can, for example, be found in Sparrow (1982).

When the presence of lateral walls is taken into account, the amplitude

equations must include extra terms expressing the variation of the amplitudes

with respect to the horizontal coordinates. An abbreviated version is the

following

∂A

∂t

= σA−lA

3

+β

∂

2

A

∂x

2

, (6.36)

which is called the Ginzburg–Landau’s equation in reference to a paper by

these authors on superconductivity, wherein an equation of the above form

was given.

154 6 Instabilities and Pattern Formation

The main important result of the above analysis is that the inﬁnite hor-

izontal plane should be covered with straight parallel rolls of inﬁnite length

and oriented randomly. The randomness of the pattern ﬁnds its origin in

the randomness of the initial disturbances in an inﬁnite plane theory. This

degeneracy may be overcome either by enclosing the ﬂuid in boxes of ﬁnite

aspect ratio or by using a grid with an imposed spacing. When the grid is

slightly heated by a lamp and placed above the upper cooled surface, it in-

duces rolls whose wavelength corresponds to the grid spacing. By increasing

the Rayleigh number, one can maintain these rolls and follow their evolution

in the course of time.

6.3.3 B´enard–Marangoni’s Surface Tension-Driven

Instability

In Sect. 6.3.2, we have shown that, in suﬃciently deep ﬂuid layers, convection

sets in when buoyancy overcomes heat dissipation and viscous forces, and that

only rolls are stable for small and moderate Rayleigh numbers. Alternatively,

in rather shallow layers with a upper surface open to air, as in B´enard’s

original experience, a regular pattern of hexagons is observed as shown in

Fig. 6.7.

It is now admitted that it is the temperature dependence of the surface

tension at interface, which is responsible for the hexagonal ﬂow pattern.

Pearson (1958) was the ﬁrst to propose a theoretical analysis on the inﬂuence

of the surface tension o assumed to depend linearly on the temperature

Fig. 6.7 B´enard’s hexagonal pattern viewed from above (from Koschmieder 1993)

6.3 Thermal Convection 155

o = o

0

+

∂o

∂T

(T −T

0

), (6.37)

wherein ∂o/∂T is usually a negative quantity. The linear analysis is similar

to that of the Rayleigh–B´enard’s problem, but the dimensionless Rayleigh

number is now replaced by the Marangoni number

Ma = −

∂o

∂T

∆T

d

ρ

0

νχ

, (6.38)

where ∆T is the temperature diﬀerence applied between the boundaries of

the ﬂuid layer. At the upper surface, the boundary conditions read now, in

dimensionless form,

W = 0, D

2

W = k

2

Ma Θ (D = d/dZ). (6.39)

The ﬁrst relation arises from the assumed non-deformability of the surface,

whereas the second one expresses the equality between the normal compo-

nent of the mechanical stress tensor and the horizontal gradient of the surface

tension (see Problem 6.8). For a stress-free boundary, the Marangoni num-

ber vanishes and one ﬁnds back the classical result D

2

W = 0. It is impor-

tant to realize that there are signiﬁcant diﬀerences between the Rayleigh

and Marangoni numbers, as the former varies as d

3

and the latter as d;

this means that surface tension-driven instability will be dominant in thin

layers (small d), whereas buoyancy eﬀects will be more important in thick

layers (large d). Remembering that in B´enard’s original experiments the

depth of the layer was about 1 mm, it is not surprising that the instabil-

ity observed by B´enard was essentially caused by surface tension forces. If

the ratio Ma/Ra →∞, the surface forces dominate the dynamics of the ﬂuid

and gravity eﬀects can be neglected. Under terrestrial conditions, Ma/Ra = 4

for a layer with a thickness d = 1 mm and Ma/Ra = 4 10

−4

for d = 10 cm.

Under microgravity conditions (g ≈ 0), Ma/Ra = 10

4

for d = 1 mm.

As an illustration, consider a ﬂuid layer (in which buoyancy eﬀects are

neglected, i.e. Ra = 0) in contact with a perfectly heat conducting bot-

tom plate and whose upper surface is adiabatically isolated but subject to a

temperature-dependent surface tension. The marginal stability curve giving

(Ma)

0

as a function of k is similar to the marginal curve of the Rayleigh–

B´enard’s problem (Pearson 1958)

(Ma)

0

=

8k

2

cosh k(k −sinh k cosh k)

k

3

cosh k −sinh

3

k

, (6.40)

and is represented in Fig. 6.8. The corresponding critical values are (Ma)

c

=

79.61 and k

c

= 1.99.

The theory has been advanced further by Scriven and Sternling (1964) to

allow for deformations of the free surface, in which case the onset of convec-

tion may be oscillatory. Nield (1964) extended Pearson’s analysis to account

for coupled buoyancy and surface tension eﬀects. The numerical results (see

156 6 Instabilities and Pattern Formation

Fig. 6.8 Marginal stability curves for coupled buoyancy and surface tension-driven

instability; the curve Ra = 0 corresponds to surface tension-driven instability

Fig. 6.8) show that the critical value (Ma)

c

decreases upon increasing Ra

but that the critical wave number remains practically unchanged. Nield was

able to derive the following linear relation between the critical Rayleigh and

Marangoni numbers

Ra

(Ra)

c

+

Ma

(Ma)

c

= 1, (6.41)

which exhibits the strong coupling between the two motors of instability;

(Ra)

c

and (Ma)

c

are the critical values corresponding to the absence of sur-

face tension gradients and gravity, respectively.

All the above theoretical considerations are based on linearized equations;

they give interesting information on the onset of convection but are silent

about the planform and other characteristics of the ﬂow; to go further a

non-linear analysis is required.

Non-linear eﬀects were considered by Cloot and Lebon (1984), who solved

the problem with a power series expansion (Shl¨ uter et al. 1965), and by

Bragard and Lebon (1993) using the amplitude method. To incorporate pos-

sible hexagonal patterns, the velocity disturbance is taken to be of the general

form

w = W(z)

_

Z(t) cos ky +Y (t) cos(

√

3/2kx) cos(ky/2)

_

, (6.42)

6.3 Thermal Convection 157

which corresponds to rolls for Y = 0 and to hexagons for Y = 2Z. After some

lengthy but rather simple calculations, Bragard and Lebon (1993) obtained

under zero-gravity conditions the following set of amplitude equations

dY

dt

= εY −γY Z −RY

3

−PY Z

2

,

dZ

dt

= εZ −

1

4

γY

2

−R

1

Z

3

−

1

2

PY

2

Z,

(6.43)

where ε = [Ma−(Ma)

c

]/(Ma)

c

and γ, P = 4R−R

1

, R, and R

1

are constant

coeﬃcients determined from the solvability condition. Observe that the sym-

metry property f(Y ) = f(−Y ) and f(Z) = f(−Z) is broken in (6.43). To

obtain a hexagonal structure, the condition Y = 2Z must be fulﬁlled and it

is directly checked that the steady versions of (6.43) possess a solution ver-

ifying this condition. By performing a linear stability analysis of the steady

solutions, it was shown (Bragard and Lebon 1993) that hexagons are stable

in the range −0.0056 < ε < 1.8. By setting Y = 2Z in the stationary solution

of (6.43), one obtains a second-order equation in Z, namely

ε −γZ −(R

1

+ 2P)Z

2

= 0, (6.44)

which is represented in Fig. 6.9 and displays an inverted bifurcation, conﬁrm-

ing the presence of a region of subcritical instability, characterized by the

occurrence of motion at ε < 0.

Experiments by Koschmieder (1993) indicate that the pattern of the sub-

critical instability consists of ill-deﬁned small-scale hexagonal cells. For the

sake of completeness, we have also reported in Fig. 6.9 the amplitude curve

for the roll solution. The solid portions of the curves refer to stable mo-

tions and the broken lines to unstable solutions. For ε < ε

−

(= −0.0056),

Fig. 6.9 Amplitude Z vs. ε: solid lines correspond to stable solutions and broken

lines correspond to unstable solutions

158 6 Instabilities and Pattern Formation

the ﬂuid remains at rest, as soon as ε is larger than ε

−

but less than zero,

subcritical stable hexagons are predicted. Moreover, hexagons are the only

stable solutions in the interval ε

−

< ε < ε

1

(= 0.53) while in the range

ε

1

< ε < ε

2

(= 1.8), both hexagonal cells and rolls are stable; ﬁnally for

ε > ε

2

, only stable rolls can be found.

A behaviour similar to B´enard–Marangoni’s instability is observed in

Rayleigh–B´enard’s problem when Boussinesq approximation is given up. In

the particular case of a temperature-dependent viscosity, it was found (Palm

1975) that the hexagonal pattern is the preferred structure, even for a ﬂuid

enclosed between two rigid plates. Experiments have also revealed that the

motion in the hexagonal platform may be either ascending or descending at

the centre of the cell: the motion is downward in the cell centres of gases and

upward in liquids. This change in circulation is associated to the property

that, in gases, the viscosity increases with temperature while it decreases in

liquids.

Other interesting aspects of buoyancy and surface tension-driven instabil-

ities have been treated in the specialized literature (e.g. Colinet et al. 2002).

Among them, let us mention the problem of ﬂuid layers submitted to hori-

zontal temperature gradients. This is important in crystal growth processes

(Davis 1987; Parmentier et al. 1993; Madruga et al. 2003), convection in

porous media, which is of interest in the mineral oil industry, thermal con-

vection in rotating systems (Busse 1978; Davis 1987), and double diﬀusive

convection under the mutual action of temperature and salinity gradients

(thermohaline convection and salt ﬁngers). All these phenomena play a cen-

tral role in geophysics, astrophysics, and oceanography. Let us also point out

the eﬀects resulting from surface deformations (Regnier et al. 2000), partic-

ularly relevant in small boxes and in presence of lateral walls, unavoidable in

practical experiments (Rosenblat et al. 1982; Dauby and Lebon 1996; Dauby

et al. 1997). B´enard’s convection has also been used to model the Earth’s

mantle motions (Turcotte 1992); moreover, it oﬀers an attractive basis for the

interpretation of the transition to turbulence (Berg´e et al. 1984). The above

list, although being not exhaustive, reﬂects the impressively wide range of

applications of Rayleigh–B´enard–Marangoni’s instability.

6.4 Taylor’s Instability

Taylor’s instability is observed in a viscous incompressible ﬂuid column con-

tained between two vertical cylinders rotating at diﬀerent angular velocities,

the temperature being assumed to remain uniform. There occurs a compe-

tition between centrifugal forces and viscosity: if we assume that the inner

cylinder moves at a higher velocity than the outer one, the ﬂuid close to

the inside wall will move outward and replace there the slow moving ﬂuid.

At small angular velocity, one has an ordinary Couette ﬂow where angular

6.4 Taylor’s Instability 159

Fig. 6.10 (a) Taylor’s vortices, (b) wavy ﬂow, and (c) turbulence (from Coles 1965)

momentum is transported from the inner to the outer cylinder by viscosity

in the form of stationary annular convective cells. When the angular veloc-

ity is gradually increased, this state becomes unstable and toroidal vortices,

also called Taylor’s vortices, are formed. At higher angular velocities, wavy

deformations appear and ﬁnally a transition to turbulent motion occurs (see

Fig. 6.10).

At low angular velocity, the motion is laminar with a velocity ﬁeld given

by

v

r

= 0, v

θ

= Ar +

B

r

, v

z

= 0, (6.45)

when expressed in cylindrical coordinates r, θ, z, where r is the radial distance

to the axis of the cylinders, θ the azimuthal angle, and z the coordinate

parallel to the vertical axis, A and B are constants to be determined from

the boundary conditions. With Ω

1

and Ω

2

the angular velocities of the inner

and outer cylinders, respectively, and R

1

and R

2

the corresponding radii, the

non-slip condition imposes the following conditions at the walls

v

θ

= R

1

Ω

1

at r = R

1

, v

θ

= R

2

Ω

2

at r = R

2

; (6.46)

this allows us to determine the values of the constants A and B, namely

A = −Ω

1

Ω

2

−µ

1 −Ω

2

, B = Ω

1

R

2

1

(1 −µ)

1 −Ω

2

, (6.47)

where Ω = Ω

2

/Ω

1

and µ = R

1

/R

2

.

We ﬁrst proceed with a linear stability analysis. Denote the velocity dis-

turbance by u

, v

, w

and the pressure disturbance by p

; assuming that

160 6 Instabilities and Pattern Formation

the disturbances are small and axisymmetric, i.e. θ independent, it is easily

checked that they obey the following linearized continuity and Navier–Stokes

equations

1

r

∂

∂r

(ru

) +

∂w

∂z

= 0, (6.48a)

∂u

∂t

−2v

θ

v

r

= −

∂

∂r

_

p

ρ

_

−ν

_

∇

2

u

−

u

r

2

_

, (6.48b)

∂v

∂t

+

_

∂v

θ

∂r

+

v

θ

r

_

u

= ν

_

∇

2

v

−

v

r

2

_

, (6.48c)

∂w

∂t

= −

∂

∂z

_

p

ρ

_

+ν∇

2

w

, (6.48d)

where ∇

2

≡ (1/r)(∂/∂r) + ∂

2

/∂r

2

+ ∂

2

/∂z

2

. The pressure disturbance p

is

easily eliminated from relations (6.48b) and (6.48d). Moreover, by suppos-

ing that the ﬂuid column is of inﬁnite length, we need only the boundary

conditions at R

1

and R

2

, namely

u

= v

= w

= 0. (6.49)

According to the normal mode technique, we seek for solutions of the form

(u

, v

, w

) = [U(r), V (r), W(r)] exp(ikz +σt). (6.50)

To simplify the calculations, we shall make the narrow-gap approximation,

which implies that the distance d = R

2

− R

1

between the cylinders is much

smaller than R

1

. This permits us to put µ = 1 and to omit 1/r compared to

∂/∂r, moreover within the small-gap limit, A + B/r

2

≈ Ω

1

and A ≈ −Ω

1

.

Substituting (6.50) in (6.48), making all the quantities dimensionless, and

proceeding by analogy with the Rayleigh–B´enard’s problem, we obtain the

following relations for the radial and azimuthal velocity amplitudes at the

marginal state σ = 0

(D

2

−k

2

)

2

U = [1 + (Ω −1)Z]V, (6.51)

(D

2

−k

2

)V = −Ta k

2

U, (6.52)

where we have introduced the notation D = d/dZ with

Z = (r −R

1

)/d (6.53)

and

Ta = 4Ω

2

1

d

4

/ν

2

, (6.54)

the relevant boundary conditions are

U = DU = V = 0 at Z = 0 and Z = 1. (6.55)

6.4 Taylor’s Instability 161

The quantity Ta is the dimensionless Taylor number, which is the ratio be-

tween the centrifugal forces and the viscous dissipation. It is worth to stress

that relations (6.51) and (6.52) are similar to (6.1.13) and (6.1.14) describing

Rayleigh–B´enard’s instability. For each particular value of Ω = Ω

2

/Ω

1

, the

eigenvalue problem (6.51) and (6.52) results in a marginal stability curve (see

Fig. 6.11), which is similar to the curves for the Rayleigh–B´enard’s problem

in the rigid–rigid conﬁguration. The minimum of the curve gives the criti-

cal value (Ta)

c

determining the minimal angular velocity at which toroidal

axisymmetric vortices will set in. For Ta < (Ta)

c

, all the disturbances are

dampened and the ﬂow is independent of the z coordinate; at Ta = (Ta)

c

,

there is a supercritical transition from Couette ﬂow to Taylor’s vortices char-

acterized by a critical wave number k

c

or a critical wavelength λ

c

= 2π/k

c

which is easily detectable on the photographs of Fig. 6.10. In the small-gap

limit, the critical values are given by

(Ta)

c

= 3, 430/(1 +Ω), k

c

= 3.12, for 0 < Ω < 1. (6.56)

The critical Taylor number is a function of Ω and it is interesting to note

that, for Ω = 1, one has (Ta)

c

= 1, 715, which is very close to the value 1,708

obtained for the critical Rayleigh number in the rigid–rigid convection case.

The same remark applies to the critical wave number whose value k

c

= 3.12

is comparable to the value k

c

= 3.117 of the Rayleigh–B´enard’s problem. It is

therefore clear that the linear stability problems for Rayleigh–B´enard’s and

Taylor’s instabilities in the narrow-gap approximation are formally identical.

Fig. 6.11 Marginal stability curve expressing the Taylor number Ta as a function

of the wave number k

162 6 Instabilities and Pattern Formation

For wide distances between the cylinders, it is useful to work with a Taylor

number (Chandrasekhar 1961), which is an explicit function of the radius

ratio. The results remain qualitatively unchanged with respect to the small-

gap approximation; the main features are that the critical Taylor number

increases with decreasing values of the radius ratio whereas the critical wave

number remains practically unchanged.

To obtain the form of the ﬂuid ﬂow after the onset of instability, a non-

linear analysis is required. It is not our purpose to enter into the details of

the non-linear developments as they are rather intricate and therefore we

advice the interested reader to refer to the specialized literature for details.

A wide discussion can be found in Drazin and Reid (1981) and Koschmieder

(1993). Starting from an amplitude equation of the form (6.28), it is found

that, in the weakly supercritical non-linear regime, the amplitude of the ﬂow

is proportional to [Ta −(Ta)

c

]

1/2

and the motion takes the form of toroidal

vortices, the interval of stable wave numbers lies, however, in a narrower band

than the one predicted by the linear approach. The stability of the Taylor’s

vortices was studied experimentally by Coles (1965). By increasing the Taylor

number, a sequence of bifurcations is displayed, which may be summarized

as follows. For Ta < (Ta)

c

, one has a pure Couette ﬂow that becomes un-

stable at the critical Taylor number at which Taylor’s vortices are observed.

At higher values of Ta, wavy vortices with a characteristic frequency are dis-

played (see Fig. 6.10b), the next step is the appearance of a quasi-periodic

regime with two characteristic frequencies. Later on, a transition towards

weak turbulence characterized by the disappearance of one of the frequencies

takes place, still further, the second characteristic frequency vanishes and a

state of full turbulence emerges (see Fig. 6.10c). One of the interests of the

study of Taylor’s instability is that one can follow the transition from lam-

inar ﬂow to turbulence in great detail through a limited number of rather

characteristic stages, it oﬀers therefore an attractive route to an exhaustive

study of turbulence at the same footing as Rayleigh–B´enard’s convection.

In the forthcoming sections, we will discuss qualitatively several examples

of patterns occurring in such various domains as chemistry, biology, oceanog-

raphy, and electricity. Unlike the presentations of the B´enard–Marangoni’s

and Taylor’s instabilities, we shall not enter into detailed mathematical for-

mulations but rather outline the main features, referring the interested reader

to specialized books and articles.

6.5 Chemical Instabilities

Dissipative patterns are not exclusive of hydrodynamics but are also found in

many other processes and particularly in chemistry. Some chemical reactions

give rise to temporal and spatial variations of the mass concentrations of

the active substances. In the course of time, some substances may undergo

periodic oscillations whereas some spatial patterns may be formed in the

6.5 Chemical Instabilities 163

reaction vessel. It should be stressed that these phenomena are underlying

the activities of life and are therefore far from being purely academic games.

The central feature governing the appearance of self-organization in chemistry

is the autocatalytic reaction: it is a reaction in which the products of some

step take part to a subsequent step. It looks like a feedback process wherein

the presence of one particular species stimulates the production of more of

that species.

A representative example of chemical reaction exhibiting dissipative pat-

terns is the Belousov–Zhabotinsky (BZ) reaction, which is discussed later on.

When chemical reactions are coupled with diﬀusion, as it occurs in spatially

inhomogeneous mixtures, the corresponding spatio-temporal instabilities are

designated as Turing’s instabilities. In a remarkable paper, Turing (1952)

showed that a steady and uniform state in a reaction–diﬀusion system might

become unstable after some control parameter has overcome some critical value.

Maintenance of non-equilibrium states for a long period demands that

chemical reagents are added to the system. But a result of such a process

is that the concentrations of the various components are made very inho-

mogeneous and it is necessary to stir the mixture with the consequence

that spatial patterns will be destroyed. Only time oscillations can be ob-

served in such stirred homogeneous reactions. As illustrations, we will dis-

cuss in the next sections the Lotka–Volterra and the Brusselator models

(Prigogine and Lefever 1968; Nicolis and Prigogine 1977). To maintain non-

equilibrium conditions in absence of stirring has for a long time been

considered as a diﬃcult problem. It is only recently that experimental meth-

ods using molecular diﬀusion have been developed which allow for the pres-

ence of both spatial and temporal patterns.

6.5.1 Temporal Organization in Spatially

Homogeneous Systems

6.5.1.1 The Lotka–Volterra Model

A simple example of temporal organization is the Lotka–Volterra model,

which is well known in ecology. It describes the growth and death of a pop-

ulation of X individuals of a species of prey and Y individuals of a species

of predators, X and Y represent for instance two populations of ﬁshes where

Y is subsisting exclusively on X. This prey–predator system is governed by

the non-linear equations

dX

dt

= X(k

1

−k

2

Y ), (6.57)

dY

dt

= −Y (k

2

−k

3

X), (6.58)

where k

1

, k

2

, and k

3

are positive constants.

164 6 Instabilities and Pattern Formation

Relations (6.57) and (6.58) admit two trivial stationary solutions X

s

=

Y

s

= 0 and two non-trivial solutions

X

s

= k

2

/k

3

, Y

s

= k

1

/k

2

. (6.59)

Their linear stability is investigated by writing X = X

s

+x

, Y = Y

s

+y

and

by substituting these expressions in (6.57) and (6.58); making use of (6.59), it

is checked that the disturbances x

and y

**obey the following linear equations
**

dx

dt

= −k

2

X

s

y

,

dy

dt

= k

3

Y

s

x

, (6.60)

and, after elimination of y

,

d

2

x

dt

2

= −k

1

k

2

x

, (6.61)

whose solution is of the form

x

= x

0

cos(ωt) with ω =

_

k

1

k

2

. (6.62)

It follows that the stationary state is not asymptotically stable in the sense

that the perturbations do not vanish in the limit of t →∞, instead they are

periodic in time with an angular frequency ω depending on the parameters

characterizing the system. This periodic behaviour is easy to interpret: when

the population of preys X is increased, the predators Y have more food

at their disposal and their number is increasing. Since Y consumes more

individuals X, their population diminishes and less food is available for the

species Y , which in turn decreases. This reduction of the number of predators

allows a larger number of preys to survive with the consequence that their

population will increase and a new cycle is initiated. The ratio of (6.58) and

(6.57) can be written as

dY

dX

= −

Y (k

2

−k

3

X)

X(k

1

−k

2

Y )

, (6.63)

which, after integration, yields the following relation

k

3

X +k

2

Y −k

2

log X −k

1

log Y = constant, (6.64)

with the constant depending on the initial conditions, and playing the same

role as total energy in classical mechanics. Clearly, (6.64) deﬁnes an inﬁnity

of trajectories in the phase space X − Y corresponding to diﬀerent initial

conditions (see Fig. 6.12).

These trajectories take the form of concentric curves surrounding the sta-

tionary state S with a period depending on the initial conditions. This be-

haviour is typical of conservative systems and has to be contrasted with limit

cycle oscillations as exhibited by dissipative systems. This gives also a strong

argument against the use of such models to describe oscillations observed in

nature, and particularly in some classes of chemical reactions.

6.5 Chemical Instabilities 165

Fig. 6.12 Phase space trajectories: the Lotka–Volterra cycles

6.5.1.2 The Brusselator

A more appropriate modelling of the oscillatory behaviour of chemical re-

actions is provided by the Brusselator (Prigogine and Lefever 1968; Nicolis

and Prigogine 1977) which consists in a simple and instructive example of

autocatalytic scheme. The sequence of chemical reactions in the Brusselator

is the following

A

k

1

−→X, (6.65a)

2X +Y

k

2

−→3X, (6.65b)

B +X

k

3

−→Y +D, (6.65c)

X

k

4

−→E, (6.65d)

the four steps are assumed to be irreversible which is achieved by taking

all reverse reaction constants equal to zero; the global reaction of the above

scheme is

A+B →D +E. (6.66)

The concentrations of the reactants A and B are maintained at a ﬁxed and

uniform non-equilibrium value, and the ﬁnal products D and E are removed

as soon as they are formed; the autocatalytic step is the second one, which

involves the intermediate species X and Y . It is assumed that the reac-

tions take place under isothermal and well stirred, i.e. spatially homogeneous

166 6 Instabilities and Pattern Formation

conditions. According to the laws of chemical kinetics, we have the following

rate equations for the species X and Y

dX

dt

= k

1

A+k

2

X

2

Y −k

3

BX −k

4

X, (6.67)

dY

dt

= −k

2

X

2

Y +k

3

BX, (6.68)

whose steady solutions are

X

s

=

k

1

k

4

A, Y

s

=

k

3

k

4

k

1

k

2

B

A

. (6.69)

To examine the stability of this solution, let us set X = X

s

+ x

and Y =

Y

s

+ y

where x

and y

**are small disturbances; it is easy to verify that the
**

perturbations are obeying a linear system of the form

d

dt

_

x

y

_

= L

_

x

y

_

, (6.70)

where the matrix L is given by

L =

⎛

⎝

k

3

B −k

4

k

2

1

k

2

k

2

4

A

2

−k

3

B −

k

2

1

k

2

k

2

4

A

2

⎞

⎠

, (6.71)

whose eigenvalues σ

1

, σ

2

, . . . determine the stability of the system. It is left

as an exercise to show that the stationary state of (6.70) becomes unstable

(Re σ > 0) when B is larger than a critical concentration given by

B

c

=

k

4

k

3

+

k

2

1

k

2

k

3

k

2

4

A

2

. (6.72)

For B < B

c

, the system remains homogeneous; whereas at B = B

c

, the

eigenvalues are purely imaginary, which leads to undamped oscillations of

the perturbations, just like in the Lotka–Volterra model; ﬁnally for B > B

c

,

the system is unstable and the concentrations X and Y undergo periodic

oscillations of the limit cycle type, independently of the initial values of X and

Y . Such oscillations have indeed been detected in the Belousov–Zhabotinsky

reaction, where fascinating geometric patterns as concentric circles and spirals

propagating through the medium have been observed. BZ reaction is basically

a catalytic oxidation by potassium bromate KBrO of a organic compound

such as malonic acid CH

2

(COOH)

2

catalysed by the cerium Ce

3+

–Ce

4+

ion

couple.

After some transitory period, the oscillatory behaviour is evidenced by the

variation of the concentration of the Ce

3+

and Ce

4+

ions (Fig. 6.13). First,

a blue colour indicating an excess of Ce

4+

is spreading into the mixture, a

few minutes later the blue colour disappears and is replaced by a red one

6.5 Chemical Instabilities 167

Fig. 6.13 Experimentally observed oscillations of Ce

4+

and Ce

3+

ions in Belousov–

Zhabotinsky reaction

that indicates an excess of Ce

3+

ions. The process goes on with a succession

of blue, red, blue colours with a perfectly regular period, which to a certain

sense constitutes a chemical clock.

When the substances necessary for the reaction are exhausted, the oscil-

lations will die out and the system goes back to its equilibrium state. In

contrast with hydrodynamics where complexity is generally characterized by

non-homogeneities in space, in chemistry even a spatially homogeneous sys-

tem may exhibit a complex behaviour in time. More recent and detailed

analyses of BZ reaction have shown the occurrence of aperiodic and even

chaotic behaviours. There are several analogies of BZ periodic behaviour in

living organisms: heart beats, circadian rhythms, and menstrual cycles are a

few examples.

6.5.2 Spatial Organization in Spatially Heterogeneous

Systems

In absence of stirring, BZ reaction exhibits some non-trivial spatial patterns,

which arise from the interplay of the chemical reaction and the diﬀusion

process. When BZ reaction is performed in a long thin vertical tube so that

the problem consists of a single spatial dimension, one observes (Fig. 6.14)

a superposition of steady horizontal bands of diﬀerent colours corresponding

to low and high concentration region. This structure is analogous to the

Rayleigh–B´enard’s and Taylor’s patterns.

6.5.3 Spatio-Temporal Patterns in Heterogeneous

Systems: Turing Structures

The great variety of patterns present in nature, both in the animate and

inanimate world, like the captivating beauty of a butterﬂy, the blobs of a

168 6 Instabilities and Pattern Formation

Fig. 6.14 Spatial structure in Belousov–Zhabotinsky reaction

leopard tail, or biological morphogenesis has been a subject of surprise and

interrogation for several generations of scientists. The question soon arises

about the mechanism behind them. In 1952, Turing proposed an answer based

on the coupling between (chemical) reactions and diﬀusion. As an example of

a Turing structure, let us still consider the Brusselator but we suppose now

that the chemical reaction takes place in a unstirred thin layer or a usual

vessel, so that spatial non-homogeneities are allowed. The basic relations are

the kinetic equations (6.67) and (6.68) to which are added the diﬀusion terms

D

x

∇

2

X and D

y

∇

2

Y , respectively, it is assumed that the diﬀusion coeﬃcients

D

x

and D

y

are constant. By repeating the analysis of Sect. 6.5.1, it is shown

that the stationary homogeneous state becomes unstable for a concentration

B larger than the critical value

B

c

=

k

4

k

3

+

k

2

1

k

2

k

3

k

2

4

A

2

+

k

4

k

3

j

2

π

2

l

2

(D

x

+D

y

), j = 0, 1, 2 . . . (6.73)

at the condition that the diﬀusion coeﬃcients are unequal, if D

x

= D

y

diﬀusion will not generate an instability, l is a characteristic length. Non-

homogeneities will begin to grow and stationary spatial patterns will emerge

in two-dimensional conﬁgurations. A rather successful reaction for observing

Turing’s patterns is the CIMA (chlorite–iodide–malonic acid) redox reaction,

which was proposed as an alternative to BZ reaction. The oscillatory and

space-forming behaviours in CIMA are made apparent through the presence

of coloured spots with a hexagonal symmetry. By changing the concentra-

tions, new patterns consisting of parallel narrow stripes are formed instead

of the spots (see Fig. 6.15).

Although it is intuitively believed that diﬀusion tends to homogenize

the concentrations, we have seen that, when coupled with an autocatalytic

6.6 Miscellaneous Examples of Pattern Formation 169

Fig. 6.15 Examples of Turing two-dimensional structures (from Vidal et al. 1994)

reaction under far from equilibrium conditions, it actually gives rise to spatial

structures. Turing’s stationary patterns are obtained when the eigenvalues of

the L matrix given by (6.71) are real; for complex conjugate eigenvalues, the

unstable disturbances are time periodic and one observes spatio-temporal

structures taking the form of propagating waves.

The existence of spatio-temporal patterns is not exclusive to ﬂuid me-

chanics and chemistry. A multitude of self-organizations has been observed

in biology and living organisms, which are the most organized and complex

examples found in the nature. It has been conjectured that most of the prop-

erties of biological systems are the result of transitions induced by far from

equilibrium conditions and destabilizing mechanisms similar to autocatalytic

reactions. Because of their complexity, these topics will not be analysed here

but to further convince the reader about the universality of pattern forma-

tion, we prefer to discuss shortly three more examples of dissipative patterns

as observed in oceanography, electricity, and materials science.

6.6 Miscellaneous Examples of Pattern Formation

As recalled earlier, one observes many kinds of pattern formation in many

diﬀerent systems. In this section, we give a concise overview of some of them.

6.6.1 Salt Fingers

In double diﬀusion convection, the ﬂow instability is due to the coupling

of two diﬀusive processes, say heat and mass transport. In the case of the

170 6 Instabilities and Pattern Formation

salt ﬁngers, which are of special interest in oceanography, the basic ﬁelds

are temperature and salinity, i.e. the concentration of salt in an aqueous

solution; in other situations, they are for example the mass concentrations of

two solutes in a ternary mixture at uniform temperature or the concentrations

of two polymers in polymeric solutions.

If we consider diﬀusion in a binary mixture submitted to a temperature

gradient, new features appear with respect to the simple Rayleigh–B´enard’s

convection. Apart from heat convection and mass diﬀusion, the Soret and

Dufour cross-eﬀects are present and should be included in the expressions

of the constitutive equations, both for the heat ﬂux q and the mass ﬂow J.

Compared to the Rayleigh–B´enard’s problem, the usual balance equations of

mass, momentum, and energy must be complemented by a balance equation

for the mass concentration of one of the constituents while the mass density

is modiﬁed as follows to account for the mass concentration

ρ = ρ

0

[1 −α(T −T

0

) +β(c −c

0

)], (6.74)

where β stands for β = ρ

−1

0

(∂ρ/∂c). The equations for the disturbances are

then linearized in the same way as in Rayleigh–B´enard’s problem, and ﬁnite

amplitude solutions have also been analysed.

As a result of the very diﬀerent values of the molecular D and heat χ

diﬀusivity coeﬃcients (in salt sea waters D/χ = 10

−2

), some puzzling phe-

nomena are occurring. Instabilities arise when a layer of cold and pure water

is lying under a layer of hot and salty water with densities being such that

the cold water is less dense than the warm water above it; convection takes

then place in the form of thin ﬁngers of up- and down-going ﬂuids (Brenner

1970; McDougall and Turner 1982).

The mechanism responsible for the onset of instability is easily understood.

Imagine that, under the action of a disturbance, a particle from the lower

fresh cold water is moving upward. As heat conductivity is much larger than

diﬀusivity, the particle takes the temperature of its neighbouring but as it is

less dense than the saltier water outside it, the particle will rise upwards under

the action of an upward buoyancy force. Likewise if a particle from the hot

salt upper layer is sinking under the action of a perturbation, it will be quickly

cooled and, becoming denser than its surroundings, it creates a downward

buoyancy force accelerating the downward motion. This example illustrates

clearly the property that concentration non-homogeneities are dangerous for

the hydrodynamic stability of mixtures when their relaxation time is much

larger than that of temperature non-homogeneities.

The resulting ﬁnite amplitude motions have been called salt ﬁngers be-

cause of their elongated structures (see Fig. 6.16). They have been observed

in a variety of laboratory experiments with heat-salt and sugar-salt mixtures

and in subtropical oceans.

6.6 Miscellaneous Examples of Pattern Formation 171

Fig. 6.16 Salt ﬁngers (from Vidal et al. 1994)

Fig. 6.17 Temperature distributions in the ballast resistor

6.6.2 Patterns in Electricity

6.6.2.1 The Ballast Resistor

Let us ﬁrst address some attention to the ballast resistor (Bedeaux et al.

1977; Pasmanter et al. 1978; Elmer 1992); it is an interesting example be-

cause it can be described by a one-dimensional model allowing for explicit

analytic treatments and, in addition, it presents useful technological aspects.

The device consists of an electrical wire traversing through a vessel of length

L ﬁlled with a gas at temperature T

G

. The control parameters are the tem-

perature T

G

and the electric current I crossing the wire. As much as the

temperature T

G

is lower than a critical value T

c

, the temperature of the wire

remains uniform but for T

G

> T

c

there is a bifurcation in the temperature

proﬁle, which is no longer homogeneous, but instead is characterized by a

peak located at the middle of the electric wire (see Fig. 6.17).

It is worth to stress that, for T

G

> T

c

, the value of the electric current I

is insensitive to the variations of the electrical potential which indicates that

the ballast resistor can be used as a current stabilizer device.

172 6 Instabilities and Pattern Formation

6.6.2.2 The Laser

When the laser is pumped only weakly, one observes that the emitted light

waves have random phases with the result that the electric ﬁeld strength

is a superposition of random waves. Above a critical pump strength, the

laser light becomes coherent (Haken 1977), meaning that the waves are now

ordered in a well-deﬁned temporal organization (Fig. 6.18).

By increasing the external pumping, a sequence of more and more compli-

cated structures is displayed, just like in hydrodynamic and chemical insta-

bilities. In particular by pumping the laser strength above a second threshold,

the continuous wave emission is transformed into ultra-short pulses.

Fig. 6.18 Laser instability: the electrical ﬁeld strength E is given as a function

of time: (a) disordered state, (b) ordered state, and (c) the same above the second

threshold

6.6.3 Dendritic Pattern Formation

Formation of dendrites, i.e. tree-like or snowﬂake-like structures as shown

in Fig. 6.19, is a much-investigated subject in the area of pattern formation.

Fig. 6.19 Dendritic xenon crystal growth (from Gollub and Langer 1999)

6.6 Miscellaneous Examples of Pattern Formation 173

Research on dendritic crystal growth has been motivated by the necessity to

better understand and control metallurgical microstructures.

The process which determines the formation of dendrites is essentially the

degree of undercooling that is the degree to which the liquid is colder than

its freezing temperature. The fundamental rate-controlling mechanism is dif-

fusion, either diﬀusion of latent heat away from the liquid–solid interface, or

diﬀusion of chemical species toward and away from this solidiﬁcation front.

These diﬀusion processes lead to shape instabilities, which trigger the forma-

tion of patterns in solidiﬁcation. In a typical sequence of events, the initially

crystalline seed immersed in its liquid phase grows out rapidly in a cascade of

branches whose tips move outwards at a given speed. These primary arms be-

come unstable against side branching and the new side branching are in turn

unstable with respect to further side branching, ending in a ﬁnal complicated

dendritic structure. The speed at which the dendrites grow, the regularity,

and the distances between the side branches determine most of the prop-

erties of the solidiﬁed material, like its response to heating and mechanical

deformation.

To summarize, we have tried in this chapter to convince the reader

of the universality of pattern-forming phenomena. We have stressed that

similar patterns are observed in apparently very diﬀerent systems, as il-

lustrated by examples drawn from hydrodynamics (Rayleigh–B´enard’s and

B´enard–Marangoni’s convections, Taylor’s vortices), chemistry (Belousov–

Zhabotinsky’s reaction and Turing’s instability), electricity (ballast resis-

tor and laser instability), and materials science (dendritic formation). Of

course, this list is far from being exhaustive and further applications have

been worked out in a great variety of areas. Figure 6.20 displays some ex-

amples like a quasi-crystalline standing-wave pattern produced by forcing a

layer of silicone oil at two frequencies (a), a standing-wave pattern of granular

material-forming stripes (b), a typical mammalian coat as the leopard (c).

Two last remarks are in form. That a great number of particles, of the

order of 10

23

, will behave in a coherent matter despite their random thermal

Fig. 6.20 Examples of patterns in quasi-crystalline pattern (a), granular material

(b), and typical leopard’s coat (c)

174 6 Instabilities and Pattern Formation

agitation is the main feature of pattern formation. As pointed out throughout

this chapter, self-organization ﬁnds its origin in two causes: non-linear dy-

namics and external non-equilibrium constraints. Fluctuations arising from

the great number of particles and their random motion are no longer damped

as in equilibrium but may be ampliﬁed with the eﬀect to drive the system

towards more and more order. This occurs when the control parameter, like

the temperature gradient in B´enard’s experiment, crosses a critical point at

which the system undergoes a transition to a new state, characterized by

regular patterns in space and/or in time.

It may also be asked why appearance of order is not in contradiction

with the second law of thermodynamics which states that the universe is

evolving towards more and more disorder. There is of course no contradiction,

because the second principle, as enounced here, refers to an isolated system

while pattern forming can only occur in closed and/or open systems with

exchange of energy and matter with the surrounding. The decrease of entropy

in individual open or closed cells is therefore consistent with the entropy

increase of the total universe and the validity of the second law is not to be

questioned.

6.7 Problems

6.1. Non-linear Landau equation. Show that the solution of the non-linear

Landau equation

d

dt

[A[

2

= 2(Re σ) [A[

2

−2l [A[

4

is given by (6.12).

6.2. Landau equation and Rayleigh–B´enard’s instability. The Landau equa-

tion describing Rayleigh–B´enard’s instability can be cast in the form dA/dt =

σA − lA

3

whose steady solution is A

s

= (σ/l)

1/2

. Expanding σ around the

critical Rayleigh number, one has σ = α[Ra − (Ra)

c

], where α is a positive

constant from which follows the well-known result

A

s

= (α/l)

1/2

[Ra −(Ra

c

)]

1/2

.

Study the stability of this steady solution by superposing to it an inﬁnites-

imally small disturbance A

**and show that the steady non-linear solution is
**

stable for Ra > (Ra)

c

.

6.3. Third-order Landau equation. Consider the following third-order Landau

equation

dA

dt

= σA+αA

2

+βA

3

,

when A > 0, σ > 0, α < 0 and β

2

> 4ασ. For suﬃciently small values of A

at t = 0, show that A tends to the equilibrium value A

e

= −σ/α +O(σ

2

) as

t →∞.

6.7 Problems 175

6.4. Rayleigh–B´enard’s instability. Consider an incompressible Boussinesq

ﬂuid layer between two rigid horizontal plates of inﬁnite extent. The two

plates are perfectly heat conducting and the ﬂuid is heated from below. (a)

Establish the amplitude equations in the case of inﬁnitesimally small distur-

bances (linear approximation). (b) Determine the marginal instability curve

Ra(k) between the dimensionless Rayleigh number Ra and wave number k.

(c) Calculate the critical values (Ra)

c

and k

c

corresponding to onset of con-

vection.

6.5. Rotating Rayleigh–B´enard’s problem. Two rigid horizontal plates extend-

ing to inﬁnity bound a thin layer of ﬂuid of thickness d. The system, subject

to gravity forces, is heated from below and is rotating around a vertical axis

with a constant angular velocity Ω. Determine the marginal curve Ra(k)

as a function of the dimensionless Taylor number Ta = 4Ω

2

d

4

/ν

2

, where ν

is the kinematic viscosity of the ﬂuid. Does rotation play a stabilizing or a

destabilizing role?

6.6. Rayleigh–B´enard’s problem with a solute. Consider a two-constituent

mixture (solvent +solute) encapsulated between two free horizontal surfaces

and subject to a vertical temperature gradient β. Denoting by c(r, t) the

concentration of the solute, assume that the density of the mixture is given

by

ρ = ρ

0

[1 +α(T −T

0

) +γ(c −c

0

)].

Neglecting the diﬀusion of the solute so that dc/dt = 0, determine the mar-

ginal curve Ra(k) when the basic reference state is at rest with a given

concentration c

r

(z) and a temperature ﬁeld T

r

= T

0

−βz.

6.7. Non-linear Rayleigh–B´enard’s instability. A thin incompressible ﬂuid

layer of thickness d is bounded by two horizontal stress-free boundaries

(Ma = 0) of inﬁnite horizontal extent. The latter are perfectly heat con-

ducting and gravity forces are acting on the ﬂuid.

(a) Show that the convective motion can be described by the following non-

linear relation

∂w

∂t

+

1

2

∂

3

w

2

∂z

3

= ∆

3

w −Ra ∆

1

w (0 < z < 1),

here w is the dimensionless vertical velocity component, Ra the Rayleigh

number and

∆ = ∆

1

+∂

2

/∂z

2

= ∂

2

/∂x

2

+∂

2

/∂y

2

+∂

2

/∂z

2

;

the corresponding boundary conditions (at z = 0 and z = 1) are

w =

∂

2

w

∂z

2

=

∂

4

w

∂z

4

= 0.

176 6 Instabilities and Pattern Formation

(b) Find the solution of the corresponding linear problem with normal mode

solutions of the form w = W(z)f(x, y) exp(σt); to be explicit, determine

the expressions of W(z) and σ and the equation satisﬁed by f(x, y).

(c) By assuming a non-linear solution of the roll type, i.e. w = Acos(kx)g(z),

show that the Landau equation associated to this problem can be written

as

dA

dt

= αA−βA

3

,

where α and β are two constants to be determined in terms of the data

Ra, (Ra)

c

, and k

c

.

6.8. Boundary condition with surface tension gradient. In dimensional form,

the kinematic boundary conditions at a horizontal surface normal to the z-

axis and subject to a surface tension gradient may be written as σ n = ∇S,

where σ is the stress tensor or more explicitly σ

xz

+ ∂S/∂x = 0 and σ

yz

+

∂S/∂y = 0. Show that, in non-dimensional form, the corresponding boundary

condition is given by (6.39) D

2

W = k

2

Maθ. Hint: After diﬀerentiating the

ﬁrst relation with respect to x, the second with respect to y, make use of

Newton’s constitutive relation σ = η[∇v +(∇v)

T

] and the continuity relation

∇ v = 0.

6.9. B´enard–Marangoni’s instability. (a) Show that, in an incompressible liq-

uid layer whose lower boundary is in contact with a rigid plate while the upper

boundary is open to air and subject to a surface tension depending linearly

on the temperature, the marginal curve relating the Marangoni number Ma

to the wave number k is given by (6.40), i.e.

(Ma)

0

=

8k

2

cosh k(k −sinh k cosh k)

k

3

cosh k −sinh

3

k

,

both boundaries are assumed to be perfectly heat conducting and gravity

acceleration is neglected. (b) Find the corresponding critical values (Ma)

c

=

79.6 and k

c

= 1.99.

6.10. B´enard–Marangoni’s instability. The same problem as in 6.9, but now

with heat transfer at the upper surface governed by Newton’s cooling law

−λ

∂T

∂z

= h(T −T

∞

),

where λ is the heat conductivity of the ﬂuid, h the heat transfer coeﬃcient,

and T

∞

the temperature of the outside world, say the laboratory. In dimen-

sionless form, the previous law reads as DΘ = −Bi Θ, (D = d/dZ) with

Bi = hd/λ the so-called Biot number. The limiting case Bi = 0 corresponds

to an adiabatically isolated surface while Bi = ∞ describes a perfectly heat

conductor. (a) Determine the dependence of Ma with respect to k and Bi. (b)

Draw the marginal instability curves for Bi = 0, 1, 10. (c) Sketch the curves

(Ma)

c

(Bi) and k

c

(Bi).

6.7 Problems 177

6.11. The Rayleigh–B´enard–Marangoni’s instability. Show that, for the cou-

pled buoyancy–surface tension-driven instability, the Ra and Ma numbers

obey the relation (6.41)

Ra

(Ra)

c

+

Ma

(Ma)

c

= 1,

where (Ra)

c

is the critical Rayleigh number without Marangoni eﬀect and

(Ma)

c

the critical Marangoni number in absence of gravity.

6.12. The Lorenz model. (a) Show that the steady solutions of (6.25) cor-

responding to supercritical convection are given (6.35) by A

1

= A

2

=

_

b(r −1) for r > 1. (b) Prove that this solution becomes unstable at

r = Pr(Pr b + 3)/(Pr − b − 1). (c) Solve numerically the Lorenz equations

for Pr = 10, b = 8/3, and r = 28 (Sparrow 1982).

6.13. Couette ﬂow between two rotating cylinders. Consider a non-viscous

ﬂuid contained between two coaxial rotating cylinders. The reference state

is stationary with u

r

= u

z

= 0, u

θ

(r) = rΩ(r), the quantity Ω(r) is an

arbitrary function of the distance r to the axis of rotation and is related to the

reference pressure by p

ref

= ρ

_

rΩ

2

(r)dr. (a) Show that the latter result is

directly obtained from the radial component of the momentum equation. (b)

Using the normal mode technique, show that, for axisymmetric disturbances

(∂/∂θ = 0), the amplitude equation is given by

(DD

∗

−k

2

)U

r

−

k

2

σ

φ(r)U

r

= 0,

where D = d/dr, D

∗

= D + 1/r, and φ(r) = (1/r

3

)d[(r

2

Ω)

2

]/dr is the so-

called Rayleigh discriminant. It is interesting to observe that the quantity

(r

2

Ω) in the Rayleigh discriminant is related to the circulation along a circle

of radius r by

_

2π

0

u

θ

(r)r dθ = 2πr

2

Ω(r).

(c) Show further that the ﬂow is stable with respect to axisymmetric dis-

turbances if φ ≥ 0. This result reﬂects the celebrated Rayleigh circulation

criterion stating that a necessary and suﬃcient condition of stability is that

the square of the circulation does not decrease anywhere.

6.14. Lotka chemical reactions. Show that the following sequence of chemical

autocatalytic reactions

A + X →2X

X + Y → 2Y

Y + B →E + D

where the concentrations of substances A and B are maintained ﬁxed, corre-

spond to the Lotka–Volterra model.

Chapter 7

Extended Irreversible Thermodynamics

Thermodynamics of Fluxes: Memory

and Non-Local Eﬀects

With this chapter, we begin a panoramic overview of non-equilibrium ther-

modynamic theories that go beyond the local equilibrium hypothesis, which

is the cornerstone of classical irreversible thermodynamics (CIT). We hope

that this presentation, covering Chaps. 7–11, will convince the reader that

non-equilibrium thermodynamics is a fully alive and modern ﬁeld of research,

combining practical motivations and conceptual questions. Indeed, such basic

topics as the deﬁnition and meaning of temperature and entropy, the formu-

lation of the second law, and its consequences on the admissible transport

equations are still open questions nowadays.

Modern technology strives towards miniaturized devices and high-frequency

processes, whose length and timescales are comparable to the mean free path

of the particles and to the internal relaxation times of the devices, thus requir-

ing extensions of the classical transport laws studied in the previous chapters.

Indeed, these laws assume an instantaneous response of the ﬂuxes to the im-

posed thermodynamic forces, whereas, actually, it takes some time for the

ﬂuxes to reach the values predicted by the classical laws. As a consequence,

when working at short timescales or high frequencies, and correspondingly

at short length scales or short wavelengths, the generalized transport laws

must include memory and non-local eﬀects. The analysis of these generalized

transport laws is one of the main topics in modern non-equilibrium ther-

modynamics, statistical mechanics, and engineering. Such transport laws are

generally not compatible with the local equilibrium hypothesis and a more

general thermodynamic framework must be looked for.

Going beyond CIT and exploring new frontiers are the driving impetus

for the development of recent non-equilibrium thermodynamic theories. In

that respect, we will successively analyse extended irreversible thermodynam-

ics (EIT), theories with internal variables, rational thermodynamics, Hamil-

tonian formulation, and mesoscopic theories.

This overview starts with EIT, because of its formal simplicity and its prox-

imity to the methods of CIT to which the reader is already acquainted. EIT

provides a macroscopic and causal description of non-equilibrium processes

179

180 7 Extended Irreversible Thermodynamics

and is based on conceptually new ideas, like the introduction of the ﬂuxes as

additional non-equilibrium independent variables, and the search for general

transport laws taking the form of evolution equations for these ﬂuxes. Such

equations will be generally obtained by considering the restrictions imposed

by the second law of thermodynamics.

To be explicit, in EIT the space 1 of state variables is formed by the union

of the space ( of classical variables like mass, momentum, energy, and com-

position, and the space T of the corresponding ﬂuxes, i.e. 1 = ( ∪ T. The

physical nature of the T-variables is diﬀerent from that of the (-variables.

The latter are slow and conserved with their behaviour governed by the clas-

sical balance laws. In contrast, the T-variables are fast and non-conserved:

they do not satisfy conservation laws and their rate of decay is generally very

short. In dilute gases, it is of the order of magnitude of the collision time

between molecules, i.e. 10

−12

s. This means that, for time intervals much

larger than this value, fast variables can be ignored. This is no longer true in

high-frequency phenomena, when the relaxation time of the ﬂuxes is compa-

rable to the inverse of the frequency of the process, or in some materials, like

polymers, dielectrics, or superﬂuids, characterized by rather large relaxation

times of the order of seconds or minutes. The independent character of the

ﬂuxes is also made evident when the mean free path of heat or charge carri-

ers becomes comparable to the dimensions of the sample, as in nano-systems.

Other motivations for elevating the ﬂuxes to the rank of variables are given

at the end of this chapter.

The domain of application of EIT enlarges the frontiers of CIT, whose

range of validity is limited to small values of the relaxation times τ of the

ﬂuxes, i.e. to small values of the Deborah number De ≡ τ/t

M

, with t

M

a

macroscopic timescale, and to small values of the Knudsen number Kn ≡ /L,

where is the mean free path and L a macroscopic length. The transport

equations derived from EIT reduce to the CIT expressions for De ¸ 1 and

Kn ¸ 1, but are applicable to describe a wider range of situations encom-

passing De > 1 and Kn > 1. Examples of situations for which De ≥ 1 are

processes where the macroscopic timescale becomes short enough to be com-

parable to the microscopic timescale, as in ultrasound propagation in dilute

gases or neutron scattering in liquids, or when the relaxation time becomes

long enough to be comparable to the macroscopic timescale, as in polymer

solutions, suspensions, superﬂuids, or superconductors. The property Kn ≥ 1

is characteristic of micro- and nano-systems as thin ﬁlms, superlattices, sub-

micronic electronic devices, porous media, and shock waves.

A simple way, although not unique, to obtain the time evolution equations

of the ﬂuxes on a macroscopic basis is to assume the existence of a generalized

entropy and to follow the same procedure as in CIT. In EIT, it is taken for

granted that there exists a non-equilibrium entropy s to which the following

properties are assigned:

• It is an additive quantity.

• It is a function of the whole set of variables s = s(1).

7.1 Heat Conduction 181

• It is a concave function of the state variables.

• Its rate of production is locally positive.

Once the expression of s is known, it is an easy matter to derive generalized

equations of state, which are of interest in the description of non-equilibrium

steady states.

The scope of this chapter is to give a general presentation of what EIT is,

how it works, and what can be expected from it. EIT has been the object of

several monographs by Jou et al. (2000, 2001), Lebon (1992) and M¨ uller and

Ruggeri (1998). For a more microscopic perspective, we refer to the books

by Eu (1992, 1998) and Luzzi et al. (2001, 2002); see also two collections of

contributions by several authors (Casas-V´ azquez et al. 1984; Sieniutycz and

Salamon 1992) or reviews by Garcia-Colin (1991, 1995) and Nettleton and

Sobolev (1995).

As an introductory example, we will study heat conduction in a rigid body

with memory eﬀects; in this problem, only the heat ﬂux is introduced as extra

variable. Afterwards, we shall discuss more complicated situations, such as

viscous ﬂuids, polymer solutions, and electric transport in microelectronic

devices, where other ﬂuxes, as the viscous pressure tensor and the electric

current, are selected as supplementary independent variables.

7.1 Heat Conduction

After outlining the shortcomings of the classical approach, we shall motivate

the choice of the heat ﬂux as independent variable in the prototype problem

of heat conduction in a non-deformable solid at rest, and we will take advan-

tage of this particular example to introduce the main tenets of the general

formalism.

7.1.1 Fourier’s vs. Cattaneo’s Law

The best-known model for heat conduction is Fourier’s law, which relates

linearly the temperature gradient ∇T to the heat ﬂux q according to

q = −λ∇T, (7.1)

where λ is the heat conductivity, depending generally on the temperature.

By substitution of (7.1) in the energy balance equation, written in absence

of source terms as

ρ

∂u

∂t

= −∇ q, (7.2)

and relating the speciﬁc internal energy u to the temperature by means of

du = c

v

dT, with c

v

being the heat capacity per unit mass at constant volume,

one obtains

ρc

v

∂T

∂t

= ∇ (λ∇T). (7.3)

182 7 Extended Irreversible Thermodynamics

The material time derivative has been replaced here by the partial time deriv-

ative because the body is at rest. From a mathematical point of view, (7.3)

is a parabolic diﬀerential equation. Although this equation is well tested for

most practical problems, it fails to describe the transient temperature ﬁeld

in situations involving short times, high frequencies, and small wavelengths.

For example, it was shown by Maurer and Thomson (1973) that, by sub-

mitting a thin slab to an intense thermal shock, its surface temperature is

300

◦

C larger than the value predicted by (7.3). The reasons for this failure

must be found in the physical statement of Fourier’s law, according to which

a sudden application of a temperature diﬀerence gives instantaneously rise to

a heat ﬂux everywhere in the system. In other terms, any temperature dis-

turbance will propagate at inﬁnite velocity. Physically, it is expected, and it

is experimentally observed, that a change in the temperature gradient should

be felt after some build-up or relaxation time, and that disturbances travel

at ﬁnite velocity. From a microscopic point of view, Fourier’s law is valid in

the collision-dominated regime, where there are many collisions amongst the

particles, but it loses its validity when one approaches the ballistic regime, in

which the dominant collisions are those of the particles with the boundaries

of the system rather than the collisions amongst particles themselves.

To eliminate these anomalies, Cattaneo (1948) proposed a damped version

of Fourier’s law by introducing a heat ﬂux relaxation term, namely

τ

∂q

∂t

= −(q +λ∇T). (7.4)

The term containing the time τ represents the heat ﬂux relaxation. When

the relaxation time τ of the heat ﬂux is negligible or when the time variation

of the heat ﬂux is slow, this equation reduces to Fourier’s law. For homoge-

neous solids, τ describes molecular-scale energy transfer by either phonons

or electrons, and it is very small, of the order of time between two successive

collisions at the microscopic level. Therefore, in most practical heat transfer

problems, inﬁnite propagation is not relevant as those parts of the signals

with inﬁnite velocity are strongly damped at room temperature. However,

when slow internal degrees of freedom are involved, as in polymers, superﬂu-

ids, porous media, or organic tissues, τ reﬂects the time required to transfer

energy between diﬀerent degrees of freedom and it may be relatively large,

of the order or larger than 1 s.

Relaxational eﬀects on heat conductors were already discussed by Maxwell

(1867), Cattaneo (1948), Vernotte (1958) and Grad (1958), but without refer-

ring to its thermodynamic implications, which will be analysed in Sect. 7.1.2

in the frame of EIT.

Assuming constant values of c

v

and λ and introducing (7.4) in (7.2) results

in the following hyperbolic equation

τ

∂

2

T

∂t

2

+

∂T

∂t

−χ∇

2

T = 0, (7.5)

7.1 Heat Conduction 183

where χ = λ/ρc

v

designates the heat diﬀusivity. Equation (7.5) is sometimes

called telegrapher’s equation because it is similar to the one describing prop-

agation of electrical signals along a wire. For small values of the time t ¸τ,

the ﬁrst term of (7.5) is dominant, so that it reduces to

τ

∂

2

T

∂t

2

= χ∇

2

T. (7.6)

This is a wave equation with a wave propagating at the velocity (χ/τ)

1/2

; it

describes a reversible process, as it is invariant with respect to time inversion.

In contrast, for timescales much longer than τ(t ¸τ), the ﬁrst term of (7.5)

is negligible and one obtains a partial diﬀerential equation of the form

∂T

∂t

= χ∇

2

T, (7.7)

which is associated with diﬀusion of heat, as shown in Chap. 2. Diﬀusion is

typically an irreversible process, as (7.7) is not invariant when t is changed

into −t. To summarize, at short times the transport equation (7.5) is re-

versible and heat propagates as a wave with a well-deﬁned speed (which

may be microscopically interpreted as a ballistic motion of heat carriers),

whereas at longer times the process becomes irreversible and heat is diﬀused

throughout the system. It is therefore clear that τ can be interpreted as the

characteristic time for the crossover between ballistic motion and the onset of

diﬀusion. In the context of chaotic deterministic systems, τ is interpreted as

the Lyapunov time beyond which predictivity is lost (Nicolis and Prigogine

1989).

The dynamical properties of (7.5) have been thoroughly analysed. By

assuming that there exists a solution of the form T(x, t) = T

0

exp[i(kx−ωt)],

where ω is a (real) frequency and k is a (complex) wave number, it is found

(see Problem 7.2) that the solution is characterized by a phase speed v

p

and

an attenuation length α, respectively, given by

v

p

=

ω

Re k

=

√

2χω

_

τω +

√

1 +τ

2

ω

2

, α =

1

Imk

=

2χ

v

p

. (7.8)

In the low-frequency limit (τω ¸1), it is found that

v

p,0

=

_

2χω, α

0

=

_

2χ/ω, (7.9)

which are the results obtained directly from Fourier’s law. In the high-

frequency limit (τω ¸1), the phase speed and attenuation length are

v

p,∞

≡ U = (χ/τ)

1/2

, α

∞

= 2(χτ)

1/2

. (7.10)

The velocity U corresponds to the so-called second sound, which is a tem-

perature wave, not to be confused with the ﬁrst sound, which is a pressure

wave with velocity c

0

=

_

(∂p/∂ρ)

s

. The value of v

p,∞

diverges when the

184 7 Extended Irreversible Thermodynamics

relaxation time vanishes, thus leading to an inﬁnite speed of propagation.

The presence of the relaxation time represents much more than a minor cor-

rection to the classical results, as in the high-frequency regime it leads to a

completely diﬀerent behaviour than that predicted by the classical Fourier’s

law. In Sect. 7.1.3, we discuss other physical consequences of (7.5). The analy-

sis of thermal waves has been the topic of much research (Joseph and Preziosi

1989; Dreyer and Struchtrup 1993; Tzou 1997; Jou et al. 2001).

The diﬀerences between parabolic and hyperbolic equations are shown in

Box 7.1. Figure 7.1 provides a qualitative comparison between the parabolic

(τ = 0) and hyperbolic (τ ,= 0) solutions of (7.5), at a short time after appli-

cation of a thermal pulse in a one-dimensional rod. According to Cattaneo’s

Box 7.1 Diﬀerences Between Hyperbolic and Parabolic Heat

Transport

As shown in Box 2.3, the response to a delta pulse of temperature is given,

according to the classical diﬀusion equation, by

∆T(x, t) =

g

0

2(πχt)

1/2

exp

_

−

x

2

4χt

_

, (7.1.1)

where g

0

is a value which depends on the intensity of the initial pulse.

The response to a delta pulse predicted by the telegrapher’s equation is

much more complicated, but has been derived in many textbooks (see, in

particular, Morse and Feshbach 1953; Jou et al. 2001; Tzou 1997). It is

found that

∆T(x, t) =

g

0

4

exp

_

−

t

2τ

__

2I

0

(ξ)Uτδ(Ut −[x[)

+

_

I

0

(ξ) +

t

τξ

dI

0

(ξ)

dξ

_

H(Ut −[x[)

_

, (7.1.2)

where I

0

(x) is the modiﬁed Bessel function of the ﬁrst kind, H(x) the

Heaviside function, U the second sound velocity given by U ≡ (χ/τ)

1/2

and ξ ≡

1

2

[(t/τ)

2

− (x/Uτ)

2

]

1/2

. The main diﬀerences with the solution

of the classical diﬀusion equation (7.1.1) are shown in Fig. 7.1. They are

essentially:

1. In the classical description, ∆T(x, t) is non-zero at any position in space

for t > 0. In contrast, ∆T(x, t) given by (7.1.2) vanishes for x > Ut, U

being the speed of the front.

2. In (7.1.2) there are two main contributions: the ﬁrst, in δ(Ut − x), de-

scribes the propagation of the initial pulse with speed U, damped by the

exponential term. The second term, in H(Ut −x), is related to the wake,

which, for suﬃciently long times Ut ¸x, tends to the diﬀusive solution

(7.1.1).

7.1 Heat Conduction 185

Fig. 7.1 Fourier (solid line) and Cattaneo (dashed line) temperature distributions

in one-dimensional rod: (a) corresponds to a given time t

1

and (b) corresponds to a

time t

2

> t

1

law, the wave propagates with a speed v

p,∞

= (χ/τ)

1/2

through the sys-

tem; at the right of the wave front, located at = v

p,∞

t = (χ/τ)

1/2

t, the

temperature perturbation is not felt. The quantity is called the penetra-

tion depth, which in Fourier’s law (τ = 0) extends to inﬁnity, making that

the disturbance is instantaneously felt throughout the system (see Fig. 7.1a).

The temperature behind the wave front will be higher than the one corre-

sponding to Fourier’s law, because the same amount of energy is conﬁned

into a smaller volume. When the time is increased, the diﬀusion of energy

over a larger volume has a damping eﬀect on the wave, and the temperature

distribution becomes closer to the Fourier prediction (Fig. 7.1b). As time in-

creases, the diﬀerence between the hyperbolic and the parabolic temperature

distributions becomes smaller and smaller.

Despite the success of Cattaneo’s model and its generalizations to describe

heat transfer at high frequencies, its thermodynamic consequences are less

known and are worth examining. First, it should be stressed that Cattaneo’s

equation is not compatible with CIT. Indeed, after substitution of Cattaneo’s

equation (7.4) in the classical expression for the entropy production as derived

in Chap. 2, namely σ

s

= q ∇T

−1

, it is found that

σ

s

=

λ

T

2

(∇T)

2

+

τ

T

2

∂q

∂t

∇T, (7.11)

which is no longer positive deﬁnite, because of the presence of the second

term of the right-hand side. The problem ﬁnds its origin in the local equi-

librium hypothesis, which must therefore be revisited. This is better seen by

examining the time evolution of entropy in an isolated rigid body. Making

use of the local equilibrium assumption, one obtains for the total entropy of

the system a non-monotonic increase, while use of EIT yields a monotonic in-

crease as it is shown by the solid curve in Fig. 7.2 (Criado-Sancho and Llebot

1993).

186 7 Extended Irreversible Thermodynamics

Fig. 7.2 The evolution of the classical entropy S

CIT

during the equilibration of an

isolated system when use is made of Cattaneo’s equation is given by the dashed curve.

The evolution of the extended entropy S

EIT

, obtained from (7.2.15), is represented

by the solid curve, which, in contrast with that of S

CIT

, increases monotonically

Instead of being monotonically increasing, the classical entropy behaves in

an oscillatory way, as shown in Box 7.2. Strictly speaking, this result is not

incompatible with the Clausius’ formulation of the second law, which states

that the entropy of the ﬁnal equilibrium state must be higher than the en-

tropy of the initial equilibrium state. However, the non-monotonic behaviour

of the entropy is in contradiction with the local equilibrium formulation of

the second law, which requires that the entropy production must be positive

everywhere at any time. Note, ﬁnally, that Cattaneo’s model does not pre-

clude that heat ﬂows from lower to higher temperature. Nevertheless, this is

not in contradiction with the Clausius’ statement of the second law, about

the impossibility of constructing a cyclic heat engine whose only eﬀect is

to transport heat from a colder to a hotter reservoir, because in (7.4) the

mentioned unusual behaviour lasts only over time intervals of the order of

the relaxation time during which Clausius’ claim is at least questionable.

Notwithstanding, this behaviour is in opposition with the local equilibrium

formulation of the second law, requiring that heat should always move from

hotter to colder regions, in absence of any other eﬀect.

Box 7.2 Evolution of Entropy in a Discrete Isolated System: CIT

vs. EIT Behaviour

We study heat transfer between two rigid bodies at temperatures T

1

and T

2

(< T

1

) put into contact. Let the total system be isolated from the

outside world and let

˙

Q be the amount of heat exchanged between the

7.1 Heat Conduction 187

two bodies per unit time. In the local equilibrium formulation, one has

S(U

1

, U

2

) = S

1

(U

1

) +S

2

(U

2

) and therefore

dS

dt

=

dS

1

dt

+

dS

2

dt

= T

−1

1

dU

1

dt

+T

−1

2

dU

2

dt

. (7.2.1)

Since the global system is isolated, dU

1

+ dU

2

= 0, and from the ﬁrst law

of thermodynamics, one has

dU

1

dt

= C

1

dT

1

dt

= −

˙

Q,

dU

2

dt

= C

2

dT

2

dt

=

˙

Q, (7.2.2)

where use has been made of the equations of state dU

i

= C

i

dT

i

, and C

i

are

the respective heat capacities of the systems.

For an isolated system, (7.2.1) represents the rate of entropy produced

inside the system, and in virtue of (7.2.2), it can be written as

dS

dt

= −(T

−1

1

−T

−1

2

)

˙

Q, (7.2.3)

or assuming that T

1

−T

2

≡ ε is small,

dS

dt

=

ε

T

2

˙

Q, (7.2.4)

where T is an intermediate temperature between T

1

and T

2

. According

to the second law of thermodynamics, (7.2.3) must be non-negative. This

implies that heat only ﬂows from the region of highest temperature to the

region of lowest temperature, according to the original Clausius’ formulation

of the second law.

The simplest hypothesis ensuring the positiveness of (7.2.4) is to as-

sume, as in Fourier’s law, that the heat ﬂux

˙

Q is proportional to the driving

force ε/T

2

, so that

˙

Q = K

ε

T

2

, (7.2.5)

with K being a positive coeﬃcient. Combining (7.2.2) with (7.2.5) one

obtains for the evolution of ε

dε

dt

= −C

−1

˙

Q, (7.2.6)

where C is deﬁned as C

−1

= C

−1

1

+ C

−1

2

. When (7.2.5) is introduced into

(7.2.6), one ﬁnds that

dε

dt

= −

K

CT

2

ε. (7.2.7)

Thus ε decays exponentially as ε = ε

0

exp[−Kt/(CT

2

)]. Now, substitution

of (7.2.4) in (7.2.3) leads to

dS

dt

= K

ε

2

T

2

≥ 0, (7.2.8)

188 7 Extended Irreversible Thermodynamics

from which follows that the entropy is a monotonically increasing function

of time.

Consider now the more general situation in which heat transfer is de-

scribed by the Cattaneo’s law

τ

d

˙

Q

dt

+

˙

Q = K

ε

T

2

, (7.2.9)

where τ is the relaxation time of

˙

Q. After combining (7.2.6) and (7.2.9),

one ﬁnds that the evolution of ε is governed by

τ

d

2

ε

dt

2

+

dε

dt

+

K

CT

2

ε = 0, (7.2.10)

which is similar to the equation of motion of a damped pendulum. The

decay of ε is no longer exponential but will exhibit oscillatory behaviour

for 4Kτ/(CT

2

) > 1. The term in dε/dt corresponds to diﬀusion of heat,

whereas the term in d

2

ε/dt

2

describes the propagation of the heat wave.

Substituting the solution of (7.2.10) in (7.2.3) with

˙

Q given by (7.2.5) leads

to an expression for the entropy S which exhibits a non-monotonic depen-

dence with respect to time, as indicated by the dashed curve in Fig. 7.2.

In EIT, the quantity

˙

Q is viewed as an independent variable of a gen-

eralized entropy S

EIT

, whose rate of variation is, up to the second order in

˙

Q,

dS

EIT

dt

= T

−1

1

dU

1

dt

+T

−1

2

dU

2

dt

−a

˙

Q

d

˙

Q

dt

(7.2.11)

instead of (7.2.1). Using the conservation of energy (7.2.2), one may write

(7.2.11) as

dS

EIT

dt

= −

_

T

−1

1

−T

−1

2

_

˙

Q−a

˙

Q

d

˙

Q

dt

, (7.2.12)

which, for small temperature diﬀerences, simpliﬁes to

dS

EIT

dt

=

_

ε

T

2

−a

d

˙

Q

dt

_

˙

Q. (7.2.13)

Assuming a linear relation between the ﬂux

˙

Q and the force, given by the

quantity inside brackets, one gets

˙

Q = K

_

ε

T

2

−a

d

˙

Q

dt

_

(7.2.14)

wherein positiveness of the entropy production requires that K > 0. Making

use of (7.2.6) and replacing (7.2.14) by (7.2.13), one obtains

dS

EIT

dt

=

C

2

K

_

dε

dt

_

2

. (7.2.15)

7.1 Heat Conduction 189

This expression is either positive or zero, but never negative. It is therefore

concluded that S

EIT

, whose evolution is plotted by the continuous curve in

Fig. 7.2, increases monotonically in the course of time and is thus compatible

with evolution equations of the Cattaneo type. For an explicit analysis of

the continuous problem, see Criado-Sancho and Llebot (1993).

7.1.2 Extended Entropy

Going back to Cattaneo’s equation (7.4), one observes that it represents truly

an evolution equation for the heat ﬂux, which can therefore be considered as

an independent variable. This is indeed the key of EIT, where it is assumed

that the entropy s = s(u, q) does depend not only on the classical variable,

namely the speciﬁc internal energy u, but also on the heat ﬂux q.

In diﬀerential form, the generalized entropy is written as follows:

ds =

_

∂s

∂u

_

du +

_

∂s

∂q

_

dq. (7.12)

In analogy with the classical theory, one can deﬁne a non-equilibrium tem-

perature θ(u, q

2

) by means of the reciprocal of (∂s/∂u)

q

. This possibility is

examined in Box 7.3. Here, we will neglect second- and higher-order contribu-

tions in q and simply identify ∂s/∂u with the inverse of the local equilibrium

temperature T(u), so that

∂s

∂u

=

1

T(u)

. (7.13)

The remaining partial derivative in (7.12) will be denoted as

_

∂s

∂q

_

= −T

−1

vα

1

(u)q, (7.14)

wherein the minus sign and the factor T

−1

v are introduced for convenience,

with v the speciﬁc volume (namely, the reciprocal of the mass density); α

1

is a

undetermined function of u alone because, like temperature, its dependence

on q

2

is assumed to be negligible. Accordingly, the ﬁnal expression of the

generalized Gibbs’ equation (7.12) is

ds = T

−1

du −T

−1

vα

1

q dq. (7.15)

From this expression and the internal energy balance law (7.2), one obtains,

for the material time derivative ˙ s of the entropy,

ρ˙ s = −T

−1

∇ q −T

−1

α

1

q ˙ q, (7.16)

190 7 Extended Irreversible Thermodynamics

or, equivalently,

ρ˙ s = −∇ (T

−1

q) +q (∇T

−1

−T

−1

α

1

q ˙ q). (7.17)

This equation can be cast in the general form of a balance equation

ρ˙ s = −∇ J

s

+σ

s

, (7.18)

with the entropy ﬂux J

s

and the (positive) entropy production σ

s

given,

respectively, by

J

s

= T

−1

q (7.19)

and

σ

s

= q (∇T

−1

−T

−1

α

1

˙ q) ≥ 0. (7.20)

Relation (7.20) has the structure of a bilinear form σ

s

= q X in the ﬂux

q and the force X, identiﬁed as the quantity within parentheses in (7.20);

it diﬀers from the classical thermodynamic force, which is simply ∇T

−1

, by

the presence of a term in the time derivative of the heat ﬂux. The simplest

way to obtain an evolution equation for q compatible with the positiveness

of σ

s

is to assume that the force X is linear in q, namely

∇T

−1

−T

−1

α

1

˙ q = µ

1

q, (7.21)

where the phenomenological coeﬃcient µ

1

may depend on u but not on q

because, as previously, third-order contributions in q are omitted. Introduc-

tion of (7.21) into (7.20) results in σ

s

= µ

1

q q ≥ 0, from which is inferred

that µ

1

> 0.

Expression (7.21) contains two non-deﬁned coeﬃcients α

1

and µ

1

, which

must be identiﬁed on physical grounds. Under steady state conditions, (7.21)

simpliﬁes to

q = −

1

µ

1

T

2

∇T. (7.22)

A comparison with Fourier’s law q = −λ∇T yields then µ

1

= (λT

2

)

−1

from

which it results λ ≥ 0. Next, by comparing (7.21) with Cattaneo’s equation

(7.4), one is led to

α

1

= τ/λT. (7.23)

Box 7.3 Non-Equilibrium Temperature vs. Local Equilibrium

Temperature

In analogy with the classical theory, we deﬁne the non-equilibrium temper-

ature θ by

θ

−1

(u, q) =

_

∂s

∂u

_

q

. (7.3.1)

Expanding θ

−1

around q = 0, and omitting terms of order higher than q

2

,

one can write

θ

−1

(u, q) = T

−1

(u) +α(u)q

2

, (7.3.2)

7.1 Heat Conduction 191

where T designates the local equilibrium temperature; the coeﬃcient α(u)

depends only on u at this order of approximation.

With the identiﬁcation (7.23), the generalized Gibbs’ equation (7.15)

takes the form

ds = θ

−1

du −

τ

ρλT

2

q dq. (7.3.3)

The integrability condition of (7.3.3) demands that

∂θ

−1

∂q

= −

∂

∂u

_

τ

ρλT

2

_

q, (7.3.4)

and, after integration,

θ

−1

(u, q) = T

−1

(u) −

1

2

∂

∂u

_

τ

ρλT

2

_

q q. (7.3.5)

This result is interesting because it provides an explicit expression of the

lowest order correction of the temperature in systems out of equilibrium.

Note that θ can be identiﬁed with T when the quantity τ/ρλT

2

is con-

stant. For an overview of the current discussions on the meaning and conse-

quences of temperature in non-equilibrium systems, see Casas-V´azquez and

Jou (2003).

With the above identiﬁcations of µ

1

and α

1

, the generalized Gibbs’ equa-

tion (7.15) takes the form

ds = T

−1

du −

τ

ρλT

2

q dq, (7.24)

wherein it is important to observe that the coeﬃcient of the new term in dq

is completely identiﬁed in terms of physical quantities, namely the relaxation

time τ and the heat conductivity λ. After integration of (7.24), the explicit

expression for the entropy outside (local) equilibrium up to second-order

terms in q is

ρs(u, q) = ρs

eq

(u) −

1

2

τ

λT

2

q q. (7.25)

The monotonic increase of extended entropy as shown in Fig. 7.2 is due to

the last term on the right-hand side of (7.25).

7.1.3 Non-Local Terms: From Collision-Dominated

Regime to Ballistic Regime

However, all is not well with the Cattaneo’s equation. Although it is quali-

tatively satisfactory, as it predicts that heat pulses and high-frequency ther-

mal waves will propagate at ﬁnite speed, some quantitativepredictions are

192 7 Extended Irreversible Thermodynamics

not suﬃciently accurate. For instance, the theory of solids predicts that

χ/τ = c

2

0

/3 in the so-called Debye approximation, with c

0

the sound velocity.

After introducing this value in (7.10), it is found that the second sound is

given by the constant value U = c

0

/

√

3, in contradiction with experiments

showing that U depends on temperature. Similarly, the kinetic theory of gases

predicts that, for monatomic gases of mass m, χ/τ = 5k

B

T/3m from which it

follows that U =

_

5k

B

T/3m, but this value deviates by more than 20% from

experimental data. One may then ask how to make EIT compatible with the

above experimental evidences. It appeared rather soon that the above dis-

crepancies ﬁnd their origin in non-local eﬀects, which were not incorporated

in Cattaneo’s original equation.

Non-local eﬀects become important when the mean free path of the heat

carriers becomes comparable to the wavelength of the external perturba-

tion or to the dimensions L of the system. In this case, the Knudsen number

Kn = /L is of the order or larger than 1 and the transport laws change from

a collision-dominated regime, i.e. a regime with many collisions amongst the

particles, to a ballistic regime, where there are few collisions amongst the par-

ticles and the predominant interactions are collisions of the particles with the

walls. The diﬀerence of behaviour of transport in both situations is remark-

able. For instance, in heat pulse experiments, one distinguishes several modes

of transport: conduction, thermal waves, and ballistic fronts, and therefore

transport equations combining these factors are desirable.

The question then arises how to include these eﬀects in the formalism.

A way out is to introduce a new extra variable, the ﬂux of the heat ﬂux,

described by a second-order tensor Q and to write, instead of the Cattaneo’s

equation (7.4), the following expression

τ

1

∂q

∂t

= −(q +λ∇T) +∇ Q. (7.26)

The tensor Q, assumed to be symmetric as conﬁrmed by the kinetic theory

of phonons (Dreyer and Struchtrup 1993), may be split in the usual form

Q = QI +

0

Q, the scalar Q being one-third of its trace and

0

Q the deviatoric

part. In a relaxational approach, the evolution equations for Q and

0

Q may

be written as

τ

0

∂Q

∂t

= −Q+β

∇ q, (7.27)

τ

2

∂

0

Q

∂t

= −

0

Q+ 2β

(∇

0

q)

s

, (7.28)

where superscript s refers to the symmetric part of the tensor. Note that,

when the relaxation times are neglected, these expressions parallel the

Newton–Stokes’ equations, with q playing the role of the ﬂuid velocity and Q

7.1 Heat Conduction 193

that of the viscous pressure tensor. This is characteristic of the hydrodynamic

regime of phonon’s ﬂow, which is indeed described by these equations.

Assuming that the relaxation times τ

0

and τ

2

are negligibly small, substi-

tuting (7.27) and (7.28) into (7.26) and restricting the analysis to the linear

approximation, one obtains, for the evolution equation of the heat ﬂux,

τ

1

∂q

∂t

= −(q +λ∇T) +β

∇

2

q +

_

β

+

1

3

β

_

∇(∇ q). (7.29)

Expression (7.29) is comparable with that obtained by Guyer and Krumhansl

(1966) from phonon kinetic theory. In this approach, heat transport is the

result of energy and momentum exchanges between massless colliding carri-

ers, called phonons. To be more explicit, the phonons undergo two types of

collisions: resistive R collisions which are collisions with defects of the lattice

and/or the boundaries of the crystal, and so-called Umklapp phonon–phonon

processes; they conserve energy but not momentum and have a characteristic

time τ

R

. The second type of collisions are referred to as normal N processes;

they conserve energy and momentum, and their characteristic time will be de-

noted by τ

N

. Starting from the Boltzmann’s equation, Guyer and Krumhansl

were able to derive the following equation for the heat ﬂux:

∂q

∂t

+

1

τ

R

q +

1

3

ρc

v

c

2

0

∇T =

1

5

c

2

0

τ

N

_

∇

2

q + 2∇(∇ q)

¸

. (7.30)

Comparison between (7.29) and (7.30) yields the following identiﬁcations

τ

1

= τ

R

,

λ

τ

R

=

1

3

ρc

v

c

2

0

, β

=

1

5

c

2

0

τ

N

τ

R

, β

=

1

3

c

2

0

τ

N

τ

R

, (7.31)

from which follows that τ

1

is associated with resistive collisions while the

non-local contributions ﬁnd their origin in the normal collisions. The result

(7.29) contains as particular cases the Fourier’s and the Cattaneo’s equations

for heat transport. Indeed:

1. For 1/τ

R

and 1/τ

N

→∞, which corresponds to very high frequency of R

and N collisions, one recovers Fourier’s law.

2. For 1/τ

R

ﬁnite, 1/τ

N

→ ∞, the frequency of normal N phonon–phonon

collisions is large compared to that of R collisions and Cattaneo’s equation

is well suited.

3. In the general case where the frequencies of R and N collisions are com-

parable, non-local eﬀects are important and Guyer–Krumhansl equation

is more adequate. It becomes also clear that the ﬂux of the heat ﬂux Q ac-

counts for the presence of the momentum-preserving N phonon–phonon

collisions.

Combining (7.29) with the energy balance equation and formulating the

ﬁnal result in the Fourier’s space (i.e. in terms of frequency ω and wave

number k), the dispersion relation between ω and k for thermal waves is (see

Problem 7.5)

194 7 Extended Irreversible Thermodynamics

iω =

−χk

2

1 + iωτ

R

+

l

2

k

2

1+iωτ

N

, (7.32)

with l

2

=

4

3

β

+β

**for longitudinal waves and l
**

2

= β

**for transverse waves.
**

For ωτ

R

¸ 1 ¸ ωτ

N

and ignoring non-local eﬀects ( = 0), (7.32) leads to

the phase velocity v

p

= (χ/τ

R

)

1/2

= c

0

/

√

3. This is the regime referred to

as second sound, and is attained when τ

N

and τ

R

are of diﬀerent orders of

magnitude. At higher frequencies, when both ωτ

N

¸ 1 and ωτ

R

¸ 1, the

phase velocity obtained from (7.32) for longitudinal waves is (Dreyer and

Struchtrup 1993, Dedeurwaerdere et al. 1994)

v

p

=

_

χ

τ

R

+

l

2

τ

R

τ

N

_

1/2

=

_

14

15

c

0

. (7.33)

This regime is typical of ballistic propagation of phonons, i.e. a ﬂow in which

phonons travel through the whole crystal without suﬀering any collision. In

fact, the theoretical result (7.33) is not fully satisfactory because experiments

assign to the ballistic speed, a value equal to c

0

.

Better agreement with experiments may be achieved when higher-order

ﬂuxes Q

1

, Q

2

, . . . , Q

N

are incorporated as independent variables in the for-

malism. Indeed, in some situations, as in diluted gases, the relaxation times

of the diﬀerent higher-order ﬂuxes are of the same order as that of the heat

ﬂux itself, in such a way that when q is taken as an independent variable,

all higher-order ﬂuxes should also be taken into account. In this way, one ob-

tains for the diﬀerent ﬂuxes a hierarchy of evolution equations for the inﬁnite

number of higher-order ﬂuxes of the form

τ

n

˙

Q

n

= −Q

n

+

l

2

n−1

τ

n−1

∇Q

n−1

−∇ Q

n+1

. (7.34)

Writing (7.34) in Fourier’s space ω k and solving the corresponding expres-

sion for the heat ﬂux, one obtains a generalized Fourier’s law with a (ω, k)-

dependent thermal conductivity

˜ q(ω, k) = −ikλ(ω, k)

˜

T(ω, k), (7.35)

where ˜ q(ω, k) and

˜

T(ω, k) are the Fourier transforms of q(r, t) and T(r, t),

respectively. It is found that the generalized heat conductivity λ(ω, k) is given

by (Dedeurwaerdere et al. 1996; Jou et al. 2001)

λ(ω, k) =

λ(T)

1 + iωτ

1

+

k

2

l

2

1

1+iωτ

2

+

k

2

l

2

2

1+iωτ

3

+

k

2

l

2

3

1+iωτ

4

+···

, (7.36)

where l

n

are characteristic lengths of the order of the mean free path, and

τ

1

, τ

2

, . . . τ

N

are the relaxation times of the respective higher-order ﬂuxes.

7.1 Heat Conduction 195

This result will be exploited in Sect. 7.1.4. Two types of truncations of the

continued-fraction expansion (7.36) have been considered. The most usual is

to assume that l

i

= 0 with i > n, for some given n. But a better approach

is to suppose that l

i

= l

n

for i > n. Asymptotic developments have been

performed, with the advantage that the whole formalism can be expressed

in terms of one single “eﬀective” relaxation time, which incorporates in a

compact way the eﬀect resulting from the introduction of an inﬁnite number

of higher-order ﬂuxes (see Jou et al. 2001).

7.1.4 Application to Steady Heat Transport

in Nano-Systems

We have mentioned that in nano-systems or submicronic electronic de-

vices, whose dimensions are comparable to the mean free path of phonons,

ballistic transport is dominant and, consequently, the validity of Fourier’s

law is questioned. Here we propose to revisit this law when the Knudsen

number Kn ¸ 1. To formulate the problem in simple terms, consider a

one-dimensional system of length L, whose opposite boundaries are at tem-

perature T and T +∆T, and transient eﬀects are ignored. Depending on the

values of Kn, the heat ﬂux q takes the following limiting forms

q = λ

∆T

L

(Kn ¸1, diﬀusive transport), (7.37a)

q = Λ∆T (Kn ¸1, ballistic transport), (7.37b)

where λ denotes the thermal conductivity and Λ is a heat conduction trans-

port coeﬃcient. In the diﬀusive limit, the heat ﬂux is proportional to the

temperature gradient, according to Fourier’s law; in contrast, in the ballistic

regime, the heat ﬂux depends only on the temperature diﬀerence, but not

on the length L of the system. The values of λ and Λ have been derived

in the kinetic theory. For a monatomic ideal gas in the diﬀusive regime, it

is found that λ =

5

2

nk

B

(k

B

T/m)

1/2

, with n the particle number density,

m the mass of the particles, k

B

the Boltzmann constant, and the mean

free path given by = (k

B

T/m)

1/2

τ, with τ the average time between suc-

cessive collisions; in the rareﬁed gas regime, the heat conduction coeﬃcient

Λ is given by Λ =

1

2

nk

B

(k

B

T/m)

1/2

. Note that the thermal conductivity

λ is proportional to , whereas Λ does not depend on it. It is also worth

to mention that computer simulations of heat transport in one-dimensional

systems suggest that in some situations Fourier’s law must be generalized in

the form q = λ(T)∆T/L

α

, with α an exponent whose value depends on the

details of the system. For instance, it is found that α = 0.63 for some an-

harmonic chains or one-dimensional gases, α = 0.5 for disordered harmonic

chains with free boundaries, or α = 1.5 for disordered harmonic chains with

ﬁxed boundaries (Lepri et al. 2003).

196 7 Extended Irreversible Thermodynamics

A simple phenomenological modelling of the transition between the diﬀu-

sive and the ballistic regimes can be achieved by introducing a heat conduc-

tivity λ(T, /L), in such a way that in whole generality

q = λ(T, /L)

∆T

L

. (7.38)

The limiting values of this generalized conductivity should be

λ(T, /L) →λ(T) for /L →0,

λ(T, /L) →

λ(T)

a

L

≡ Λ(T)L for /L →∞,

where a is a constant depending on the system.

This may be achieved, for instance, through the continued-fraction ap-

proach (7.36), which in the steady state (ω

.

= 0), allows us to deﬁne a k-

dependent thermal conductivity λ(T, k). Since the system is characterized by

one single length scale, the length L, it seems natural to identify k with 2π/L.

This yields an expression for λ(T, /L), which in the simplest case where all

the l

n

take the same value, for instance, l

2

n

=

1

4

2

, and using the asymptotic

expression corresponding to (7.36) (Problem 7.6) yields

λ(T, /L) =

λ(T)L

2

2π

2

2

_

_

1 + 4(π/L)

2

_

1/2

−1

_

, (7.39)

(see Jou et al. 2005). It is directly checked that (7.39) reduces to λ(T) for

/L → 0 and to λ(T)(L/) for /L → ∞; these are the required asymp-

totic behaviours. These considerations conﬁrm that EIT provides a consis-

tent modelling for heat transfer processes taking place not only at short times

but also at micro- and nano-scales. Despite its simplicity, (7.39) satisfactorily

ﬁts experimental results in silicon thin layers and nano-wires (Alvarez and

Jou 2007). Another possibility, which has been explored recently (Chen 2004;

Lebon et al. 2006b), is to split the heat ﬂux into a diﬀusive and a ballistic

component, the latter being associated with fast particles and a long mean

free path, the former with slow particles and a short mean free path.

7.2 One-Component Viscous Heat Conducting Fluids

Up to now, we have only considered heat transfer in rigid bodies at rest. Let

us now proceed to the more general case of a compressible one-component

isotropic viscous ﬂuid in motion subject to a temperature gradient. According

to EIT, the space of the thermodynamic state variables is the union of the

7.2 One-Component Viscous Heat Conducting Fluids 197

classical one (internal energy u and speciﬁc volume v) and the space of the

ﬂuxes, namely the heat ﬂux q, the bulk viscous pressure p

v

, and the shear

viscous pressure

0

P

v

. The generalized Gibbs’ equation takes the form

ds = T

−1

du +T

−1

p dv −T

−1

vα

0

p

v

dp

v

−T

−1

vα

1

q dq −T

−1

vα

2

0

P

v

: d

0

P

v

,

(7.40)

where we have approximated the non-equilibrium temperature by the local

equilibrium temperature (see Box 7.3), similarly the non-equilibrium pressure

has been identiﬁed with the equilibrium pressure p, the coeﬃcients α

0

.

, α

1

,

and α

2

are unknown scalar functions of u and v, their dependence with respect

to the ﬂuxes is assumed to be negligible. Compared to CIT, the last three

terms are new. Our objective is to determine the evolution equations for the

ﬂuxes satisfying the thermodynamic restrictions imposed by the second law.

We shall proceed in parallel with the procedure followed in Chap. 2, which

implies that we ﬁrst determine the expression of the entropy production σ

s

.

To this end, we combine (7.40) with the expressions of ˙ u and ˙ v as derived

from the balance laws of mass (2.16) and energy (2.18). Omitting the energy

source term, it is easily found that

ρ˙ s = −T

−1

∇ q −T

−1

p

v

∇ v −T

−1

0

P

v

:

0

V−T

−1

α

0

p

v

˙ p

v

−T

−1

α

1

q ˙ q −T

−1

α

2

0

P

v

: (

0

P

v

)

•

. (7.41)

Before deriving the entropy production

σ

s

= ρ˙ s +∇ J

s

, (7.42)

we need an expression for the entropy ﬂux J

s

. For isotropic systems, the

most general vector depending on u, v, q,

0

P

v

, and p

v

is, up to second order

in the ﬂuxes,

J

s

= T

−1

q +β

p

v

q +β

0

P

v

q. (7.43)

We assume for simplicity that the coeﬃcients β

and β

are constant; (7.43)

is supported by the kinetic theory of gases and may be viewed as a gener-

alization of the classical result J

s

= q/T. The entropy production is easily

derived from (7.42) by replacing ρ˙ s and J

s

, respectively, by their expressions

in (7.41) and (7.43). The ﬁnal result is

σ

s

= q (∇T

−1

+β

∇

0

P

v

+β

∇p

v

−T

−1

α

1

˙ q)

+p

v

(−T

−1

∇ v −T

−1

α

0

˙ p

v

+β

∇ q)

+

0

P

v

: [−T

−1

0

V−T

−1

α

2

(

0

P

v

)

•

+β

(∇

0

q)

s

]. (7.44)

198 7 Extended Irreversible Thermodynamics

Note that (7.44) has the structure of a bilinear form

σ

s

= q X

1

+p

v

X

0

+

0

P

v

:

0

X2

≥ 0, (7.45)

consisting of a sum of products of the ﬂuxes q, p

v

, and

0

P

v

and their conjugate

generalized forces X

1

, X

0

, and

0

X2

. The latter are the respective terms be-

tween the brackets. They are similar to the expressions obtained in CIT, but

they contain additional terms depending on the time and space derivatives

of the ﬂuxes.

The simplest ﬂux–force relations ensuring the positiveness of σ

s

are

X

1

= µ

1

q, X

0

= µ

0

p

v

,

0

X2

= µ

2

0

P

v

, (7.46)

with µ

0

≥ 0, µ

1

≥ 0, and µ

2

≥ 0. Writing explicitly the expressions for the

generalized forces, one obtains in the linear approximation the following set

of evolution equations for the ﬂuxes:

∇T

−1

−T

−1

α

1

˙ q = µ

1

q −β

∇

0

P

v

−β

∇p

v

, (7.47)

−T

−1

∇ v −T

−1

α

0

˙ p

v

= µ

0

p

v

−β

∇ q, (7.48)

−T

−1

0

V−T

−1

α

2

(

0

P

v

)

•

= µ

2

0

P

v

−β

(∇

0

q)

s

. (7.49)

Note that in (7.49) (

0

P

v

)

•

stands for the time derivative of

0

P

v

.

The main features issued from the above thermodynamic formalism are:

• The positiveness of the coeﬃcients µ

0

≥ 0, µ

1

≥ 0, and µ

2

≥ 0.

• The equality of the cross-coeﬃcients relates q with ∇

0

P

v

and

0

P

v

with

(∇

0

q)

s

on the one side, q with ∇p

v

and p

v

with ∇q on the other side. This

is conﬁrmed by the kinetic theory, and belongs to a class of higher-order

Onsager’s relations.

• The coeﬃcients β

and β

**appearing in the second-order terms of the
**

entropy ﬂux (7.43) are the same as the coeﬃcients of the cross-terms

in the evolution equations (7.47)–(7.49). This result is also in agree-

ment with the kinetic theory of gases. These coeﬃcients describe the cou-

pling between thermal and mechanical eﬀects; setting β

= β

= 0 means

absence of coupling in (7.47)–(7.49) and similarly in (7.43) of J

s

, which

boils down to the classical result J

s

= q/T.

To identify the coeﬃcients appearing in (7.47)–(7.49) in physical terms,

consider the particular case where the space derivatives of the ﬂuxes are

negligible. Equations (7.47)–(7.49) then reduce to

∇T

−1

−T

−1

α

1

˙ q = (λT

2

)

−1

q, (7.50)

−T

−1

∇ v −T

−1

α

0

˙ p

v

= (ζT)

−1

p

v

, (7.51)

−T

−1

0

V−T

−1

α

2

(

0

P

v

)

·

= (2ηT)

−1

0

P

v

. (7.52)

7.2 One-Component Viscous Heat Conducting Fluids 199

By setting

α

1

= τ

1

(λT)

−1

, α

0

= τ

0

ζ

−1

, α

2

= τ

2

(2η)

−1

, (7.53)

µ

1

= (λT

2

)

−1

, µ

0

= (ζT)

−1

, µ

2

= (2ηT)

−1

, (7.54)

(7.50)–(7.52) can be identiﬁed with the so-called Maxwell–Cattaneo’s laws

τ

1

˙ q +q = −λ∇T, (7.55)

τ

0

˙ p

v

+p

v

= −ζ∇ v, (7.56)

τ

2

(

0

P

v

)

•

+

0

P

v

= −2η

0

V, (7.57)

with λ > 0, ζ > 0, and η > 0 being the positive thermal conductivity, bulk

viscosity, and shear viscosity, respectively, and where τ

1

, τ

0

, and τ

2

are the

relaxation times of the respective ﬂuxes. In terms of λ, ζ, η, and the relaxation

times τ

1

, τ

0

, and τ

2

, the linearized evolution equations (7.50)–(7.52) take the

following form:

τ

1

˙ q = −(q +λ∇T) +β

λT

2

∇

0

P

v

+β

λT

2

∇p

v

, (7.58)

τ

0

˙ p

v

= −(p

v

+ζ∇ v) +β

ζT∇ q, (7.59)

τ

2

(

0

P

v

)

•

= −(

0

P

v

+ 2η

0

V) + 2β

ηT(∇

0

q)

s

. (7.60)

Extension of the present description to the problem of thermodiﬀusion in

multi-component ﬂuids has been achieved by Lebon et al. (2003). Despite in

ordinary liquids, the relaxation times are small, of the order of 10

−12

s, their

eﬀects may be observed in neutron scattering experiments; in contrast, in

dilute gases or polymer solutions, the relaxation times may be rather large

and directly perceptible in light scattering experiments and in ultrasound

propagation. To illustrate the above analysis, let us mention that Carrassi and

Morro (1972) studied the problem of ultrasound propagation in monatomic

gases and compared the results provided by Maxwell–Cattaneo’s equations

(7.55)–(7.57) and the classical Fourier–Newton–Stokes’ laws.

In Table 7.1, the numerical values of the ultrasound phase velocities c

0

/v

p

(c

0

designating the sound speed) vs. the non-dimensional mean free path

of the particles are reported. It is observed that the classical theory devi-

ates appreciably from the experimental results as is increased, whereas the

Maxwell–Cattaneo’s theory agrees fairly well with the experimental data.

Table 7.1 Numerical values of c

0

/v

p

as a function of (Carrassi and Morro 1972)

0.25 0.50 1.00 2.00 4.00 7.00

(c

0

/v

p

)

(Fourier–Newton–Stokes)

0.40 0.26 0.19 0.13 0.10 0.07

(c

0

/v

p

)

(Maxwell–Cattaneo)

0.52 0.43 0.44 0.47 0.48 0.49

(c

0

/v

p

)

(experimental)

0.51 0.46 0.50 0.46 0.46 0.46

200 7 Extended Irreversible Thermodynamics

Taking into account the identiﬁcations (7.53), the explicit expression of

the entropy, after integration of (7.40), is

s

EIT

= s

eq

(u, v) −

τ

1

v

2λT

2

q q −

τ

0

v

2ζT

p

v

p

v

−

τ

2

v

4ηT

0

P

v

:

0

P

v

. (7.61)

This expression reduces to the local equilibrium entropy s

eq

(u, v) for zero

values of the ﬂuxes. Since the entropy must be a maximum at equilibrium,

it follows that the relaxation times must be positive. Indeed, stability of

equilibrium demands that entropy be a concave function, which implies that

the second-order derivatives of s

EIT

with respect to its state variables are

negative; in particular,

∂

2

s

EIT

∂q ∂q

= −

vτ

1

λT

2

< 0,

∂

2

s

EIT

∂

0

P

v

: ∂

0

P

v

= −

vτ

2

2ηT

2

< 0,

∂

2

s

EIT

(∂p

v

)

2

= −

vτ

0

ζT

2

< 0.

(7.62)

Since the transport coeﬃcients λ, ζ, and η must be positive as a consequence

of the second law, it turns out from (7.62) that the relaxation times are pos-

itive, for stability reasons. The property of concavity of entropy is equivalent

to the requirement that the ﬁeld equations constitute a hyperbolic set. Hy-

perbolicity of evolution equations is characteristic of EIT and it is sometimes

imposed from the start as in Rational Extended Thermodynamics (M¨ uller

and Ruggeri 1998).

7.3 Rheological Fluids

Section 7.2 is dedicated to general considerations about the EIT description

of viscous ﬂuid ﬂows. As mentioned above, for most ordinary ﬂuids, the re-

laxation times are generally very small so that for main problems the role of

relaxation eﬀects is minute and can be omitted. However, this is no longer true

with rheological ﬂuids, like polymer solutions, because the relaxation times

of the macromolecules are much longer than for small molecules, and they

may be of the order of 1 s and even larger. The study of rheological ﬂuids has

been the concern of several thermodynamic approaches like rational ther-

modynamics, internal variables theories, and Hamiltonian formalisms (see

Chaps. 8–10). The main diﬀerence between EIT and other theories is that

in the former the shear viscous pressure is selected as variable, whereas in

other theories, variables related with the internal structure of the ﬂuid, as

for instance, the so-called conﬁguration tensor, are preferred. The EIT vari-

ables are especially useful in macroscopic analyses while internal variables

are generally more suitable for a microscopic understanding.

In the simple Maxwell model, which is a particular case of (7.60), the

viscous pressure tensor obeys the evolution equation

dP

v

dt

= −

1

τ

P

v

−2

η

τ

V. (7.63)

7.3 Rheological Fluids 201

This model captures the essential idea of viscoelastic models: the response to

slow perturbations is that of an ordinary Newtonian viscous ﬂuid, whereas

for fast perturbations, with a characteristic time t of the order of the re-

laxation time τ or less, it behaves as an elastic solid. However, the mater-

ial time derivative introduced in (7.63) is not very satisfactory, neither for

practical predictions nor from a theoretical viewpoint, because it does not

satisfy the axiom of frame-indiﬀerence (see Chap. 9). This has motivated to

replace (7.63) by the so-called upper-convected Maxwell model

dP

v

dt

−(∇v)

T

P

v

−P

v

(∇v) = −

1

τ

P

v

−2

η

τ

V. (7.64)

In a steady pure shear ﬂow corresponding to velocity components v =

( ˙ γy, 0, 0), with ˙ γ the shear rate, the upper-convected Maxwell model (7.64)

reads as

P

v

=

⎛

⎝

−2τη ˙ γ

2

−η ˙ γ 0

−η ˙ γ 0 0

0 0 0

⎞

⎠

. (7.65)

In contrast with the original Maxwell model, element P

v

11

of tensor (7.65)

contains a non-vanishing contribution, corresponding to the so-called normal

stresses. The upper-convected model agrees rather satisfactorily with a wide

variety of experimental data.

In (7.63) and (7.64), we have introduced one single relaxation time but

in many cases it is much more realistic to consider P

v

as a sum of several

independent contributions, i.e. P

v

=

j

P

v

j

with each P

v

j

obeying a linear

evolution equation such as (7.63) or (7.64), characterized by its own viscosity

η

j

and relaxation time τ

j

. These independent contributions arise from the

diﬀerent internal degrees of freedom of the macromolecules. In this case, the

“extended” entropy should be written as

s(u, v, P

v

i

) = s

eq

(u, v) −

v

4T

i

τ

i

η

i

P

v

i

: P

v

i

(7.66)

and the corresponding model is known as the generalized Maxwell model.

In the above descriptions, the viscosity was supposed to be independent of

the shear rate. However, there exists a wide class of so-called non-Newtonian

ﬂuids characterized by shear rate-dependent viscometric functions, like the

viscous coeﬃcients. Such a topic is treated at full length in specialized works

on rheology and will not be discussed here any more.

The study of polymer solutions is often focused on the search of consti-

tutive laws for the viscous pressure tensor. One of the advantages of EIT is

to establish a connection between such constitutive equations and the non-

equilibrium equations of state derived directly by diﬀerentiating the expres-

sion of the extended entropy. These state equations are determinant in the

study of ﬂowing polymer solutions, which is important in engineering, since

most of polymer processing take place under motion. The phase diagrams

established for equilibrium situations cannot be trusted in the presence of

202 7 Extended Irreversible Thermodynamics

Fig. 7.3 Phase diagram (temperature T vs. volume fraction φ) of a polymer solution

under shear ﬂow. This is a binary solution of polymer polystyrene in dioctylphtalate

solvent for several values of P

v

12

(expressed in Nm

−2

). The dashed curve is the equi-

librium spinodal line (corresponding to a vanishing viscous pressure)

ﬂows as the latter may enhance or reduce the solubility of the polymer and

the conditions under which phase separation occurs.

This explains why many eﬀorts have been devoted to the study of ﬂow-

induced changes in polymer solutions (Jou et al. 2000, 2001; Onuki 1997,

2002). Classical local equilibrium thermodynamics is clearly not a good

candidate because the equations of state should incorporate explicitly the in-

ﬂuence of the ﬂow. Moreover, the equilibrium thermodynamic stability con-

ditions cannot be extrapolated to non-equilibrium steady states, unless a

justiﬁcation based on dynamic arguments is provided. According to EIT, the

chemical potential will explicitly depend on the thermodynamic ﬂuxes, here

the viscous pressure. It follows that the physico-chemical properties related

to the chemical potential – as for instance, solubility, chemical reactions,

phase diagrams, and so on – will depend on the viscous pressure, and will

be diﬀerent from those obtained in the framework of local equilibrium ther-

modynamics. This is indeed observed in the practice, see Fig. 7.3, where it is

shown that the critical temperature of phase change predicted by the equilib-

rium theory (281.4 K), as developed by Flory and Huggins, is shifted towards

higher values under the action of shear ﬂow. The corrections are far from

being negligible when the shear is increased.

7.4 Microelectronic Devices

The classical thermodynamic theory of electric transport has been exam-

ined in Chap. 3. Here, we brieﬂy discuss the EIT contribution to the study

of charge transport in submicronic electronic devices. Although the carrier

7.4 Microelectronic Devices 203

transport can always be described by means of the Boltzmann’s equation, to

solve it is a very diﬃcult task and, furthermore, it contains more information

than needed in practical applications. The common attitude is to consider a

reduced number of variables (usually expressed as moments of the distribu-

tion function), which are directly related with density, charge ﬂux, internal

energy, energy ﬂux, and so on, and which are measurable and controllable

variables, instead of the full distribution function. This kind of approach is

referred to as a hydrodynamic model and EIT is very helpful in determining

which truncations among the hierarchy of evolution equations are compatible

with thermodynamics.

Before considering microelectronic systems, let us ﬁrst study electric con-

duction in a rigid metallic sample. We assume that the electric current is

due to the motion of electrons with respect to the lattice. In CIT, the inde-

pendent variables are selected as being the speciﬁc internal u and the charge

per unit mass, z

e

; in EIT, the electric current i is selected as an additional

independent variable. For the system under study, the balance equations of

charge and internal energy may be written as

ρ ˙ z

e

= −∇ i , (7.67)

ρ ˙ u = −∇ q +i E, (7.68)

with E the electric ﬁeld and i E the Joule heating term. Ignoring heat

transport (q = 0) for the moment, the generalized Gibbs’ equation takes the

form

ds = T

−1

du −T

−1

µ

e

dz

e

−αi di , (7.69)

with µ

e

being the chemical potential of electrons and α a phenomenologi-

cal coeﬃcient independent of i . By following the same procedure as in the

previous sections, it is easily checked that the evolution equation for i is

τ

e

di

dt

= −(i −σ

e

E

), (7.70)

where E

= E−T∇(T

−1

µ

e

), τ

e

is the relaxation time, and σ

e

is the electrical

conductivity, provided that α in (7.69) is identiﬁed as α = τ

e

(ρσ

e

T)

−1

. The

generalized entropy is now given by

ρs = ρs

eq

−

τ

e

2σ

e

T

i i . (7.71)

Equation (7.70), i.e. a generalization of Ohm’s law i = σ

e

E, is often used

in plasma physics and in the analysis of high-frequency currents but without

any reference to its thermodynamic context.

A challenging application is the study of charge transport in submicronic

semiconductor devices for its consequences on the optimization of their func-

tioning and design. The evolution equations for the moments are directly

obtained from the Boltzmann’s equation. Depending on the choice of vari-

ables and the level at which the hierarchy is truncated, one obtains diﬀerent

204 7 Extended Irreversible Thermodynamics

hydrodynamic models. A simple one is the so-called drift-diﬀusion model

(H¨ ansch 1991), where the independent variables are the number density of

electrons and holes, but not their energies. More sophisticated is the approach

of Baccarani–Wordeman, wherein the energy of electrons and holes is taken

as independent variables, but not the heat ﬂux, assumed to be given by the

Fourier’s law. To optimize the description, a sound analysis of other possible

truncations is highly desirable. Application of EIT to submicronic devices

has been performed in recent works (Anile and Muscato 1995, Anile et al.

2003) wherein the energy ﬂux rather than the electric ﬂux is raised to the

level of independent variables. We will not enter furthermore into the details

of the development as they are essentially based on Boltzmann’s equation for

charged particles, which is outside the scope of this book. Let us simply add

that a way to check the quality of the truncation is to compare the predictions

of the hydrodynamic models with Monte Carlo simulations.

In particular, for a n

+

−n−n

+

silicon diode (Fig. 7.4) at room temperature,

the EIT model of Anile et al. (2003) provides results, which are in good

agreement with Monte Carlo simulations, as reﬂected by Fig. 7.5. Compared

to a Monte Carlo simulation, the advantage of a hydrodynamic model is its

much more economical cost with regard to the computing time consumption.

0.1mm 0.4 m m 0.1 m m

Fig. 7.4 A n

+

−n−n

+

silicon diode. The doping density in the region n

+

is higher

than in the region n

Fig. 7.5 Velocity proﬁles in the n

+

−n−n

+

silicon diode obtained, respectively, by

Monte Carlo simulations (dotted line) and the hydrodynamical model of Anile and

Pennisi (1992) based on EIT

7.5 Final Comments and Perspectives 205

7.5 Final Comments and Perspectives

To shed further light on the scope and perspectives of EIT, some general

comments are in form:

1. In EIT, the state variables are the classical hydrodynamic ﬁelds sup-

plemented by the ﬂuxes provided by the balance laws, i.e. the ﬂuxes

of mass, momentum, energy, electric charge, and so on. This attitude

is motivated by the fact that these dissipative ﬂuxes are typically non-

equilibrium variables vanishing at equilibrium. The choice of ﬂuxes is

natural as the only accessibility to a given system is through its bound-

aries. Moreover in processes characterized by high frequencies or systems

with large relaxation times (polymers, superﬂuids, etc.) or short-scale di-

mensions (nano- and microelectronic devices), the ﬂuxes lose their status

of fast and negligible variables and ﬁnd naturally their place among the

set of state variables. Other ﬁelds where the ﬂuxes may play a leading

part are relativity, cosmology, traﬃc control (ﬂux of cars), economy (ﬂux

of money), and world wide web (ﬂux of information). The choice of the

ﬂuxes as variables ﬁnds its roots in the kinetic theory of gases. Indeed, it

amounts to selecting as variables the higher-order moments of the velocity

distribution function; in particular, taking the heat ﬂux and the pressure

tensor as variables is suggested by Grad’s thirteen-moment theory (1958),

which therefore provides the natural basis for the development of EIT

(Lebon et al. 1992). The main consequence of elevating the ﬂuxes to the

rank of variables is that the phenomenological relations of the classical

approach (CIT) are replaced by ﬁrst-order time evolution equations of

Maxwell–Cattaneo type. In EIT, the ﬁeld equations are hyperbolic; note,

however, that this property may not be satisﬁed in the whole space of

state variables, especially in the non-linear regime (M¨ uller and Ruggeri

1998; Jou et al. 2001). In CIT, the balance laws are parabolic of the dif-

fusion type with the consequence that signals move at inﬁnite velocity.

EIT can be viewed as a generalization of CIT by including inertia in the

transport equations.

2. The space of the extra variables is not generally restricted to the above

ordinary dissipative ﬂuxes. For instance, to cope with the complexity of

some fast non-equilibrium processes and/or non-local eﬀects as in nano-

systems, it is necessary to introduce higher-order ﬂuxes, such as the ﬂuxes

of the ﬂuxes, as done in Sect. 7.1.3. Moreover, it is conceivable that ﬂuxes

may be split into several independent contributions, each with its own

evolution equation, as in non-ideal gases (Jou et al. 2001) and polymers

(see Sect. 7.3). In some problems, like those involving shock waves (Valenti

et al. 2002), it may be more convenient to use as variables combinations

of ﬂuxes and transport coeﬃcients.

3. Practically, it is not an easy task to evaluate the ﬂuxes at each instant

of time and at every point in space. Nevertheless, for several problems of

practical interest, such as heat wave propagation, the ﬂuxes are eliminated

206 7 Extended Irreversible Thermodynamics

from the ﬁnal equations. Although the corresponding dispersion relations

may still contain the whole set of parameters appearing in the evolution

equations of the ﬂuxes, like the relaxation times, the latter may however

be evaluated by measuring the wave speed, its attenuation, or shock prop-

erties. A direct measurement of the ﬂuxes is therefore not an untwisted

condition to check the bases and performances of EIT.

4. There are several reasons that make preferable to select the ﬂuxes rather

than the gradients of the classical variables (for instance, temperature

gradient or velocity gradient) as independent variables. (a) The ﬂuxes are

associated with well-deﬁned microscopic operators, and as such allow for

a more direct comparison with non-equilibrium statistical mechanics and

the kinetic theory. (b) The ﬂuxes are generally characterized by short

relaxation times and therefore are more adequate than the gradients for

describing fast processes. Of course, for slow or steady phenomena, the use

of both sets of variables is equivalent because under these conditions the

former ones are directly related to the latter. (c) Expressing the entropy

in terms of the ﬂuxes oﬀers the opportunity to generalize the classical

theory of ﬂuctuations and to evaluate the coeﬃcients of the non-classical

part of the entropy as will be shown in Chap. 11. This would not be

possible by taking the gradients as variables. (d) Finally, the selection of

the gradients as extra variables leads to the presence of divergent terms

in the formulation of constitutive equations, a well-known result in the

kinetic theory.

5. EIT provides a strong connection between thermodynamics and dynam-

ics. In EIT, the ﬂuxes are no longer considered as mere control parameters

but as independent variables. The fact that EIT makes a connection be-

tween dynamics and thermodynamics should be underlined. EIT enlarges

the range of applicability of non-equilibrium thermodynamics to a vast do-

main of phenomena where memory, non-local, and non-linear eﬀects are

relevant. Many of them are ﬁnding increasing application in technology,

which, in turn, enlarges the experimental possibilities for the observation

of non-classical eﬀects in a wider range of non-equilibrium situations.

6. It should also be underlined that EIT is closer to Onsager’s original con-

ceptualization than CIT. Indeed, according to Onsager, the ﬂuxes are

deﬁned as the time derivative of the state variables a

α

, and the forces are

given by the derivatives of the entropy with respect to the a

α

s

J

α

=

da

α

dt

, X

α

=

∂s

∂a

α

. (7.72)

Following Onsager, the time evolution equations of the a

α

s are obtained

by assuming linear relations between ﬂuxes and forces

da

α

dt

=

β

L

αβ

X

β

. (7.73)

Now, the ﬂuxes and forces of CIT are completely unrelated to Onsager’s

interpretation; clearly, the heat ﬂux and the pressure tensor are not time

7.5 Final Comments and Perspectives 207

derivatives of state variables, similarly, the forces ∇T and V, widely used

in CIT, cannot be considered as derivatives of s with respect to the vari-

ables a

α

. Turning now back to EIT, one can deﬁne generalized ﬂuxes J

α

and forces X

α

, respectively, by

J

q

=

dq

dt

, X

q

=

∂s

∂q

= αq, (7.74)

where q is the heat ﬂux or any other ﬂux variable and α is a phenomeno-

logical coeﬃcient. Assuming now a linear ﬂux–force relation J

q

= LX

q

,

with L = 1/ατ, one obtains an evolution equation for the state variables

q of the form

dq

dt

=

1

τ

(q

ss

−q), (7.75)

where q

ss

≡ −λ∇T is the classical Fourier steady state value of q. After

recognizing in (7.75) a Cattaneo-type relation, it is clear that the structure

of EIT is closer to Onsager’s point of view than that of CIT. Moreover,

by transposing Onsager’s arguments, it can be shown that the phenom-

enological coeﬃcient L is symmetric (Lebon et al. 1992; Jou et al. 2001).

7. Extended irreversible thermodynamics is the ﬁrst thermodynamic theory

which proposes an explicit expression for non-equilibrium entropy and

temperature. In most theories, this problem is even not evoked or the

temperature and entropy are selected as their equilibrium values, as for

instance in the kinetic theory of gases.

To summarize, the motivations behind the formulation of EIT were the

following:

• To go beyond the local equilibrium hypothesis

• To avoid the paradox of propagation of signals with an inﬁnite velocity

• To generalize the Fourier, Fick, Stokes, and Newton laws by including:

– Memory eﬀects (fast processes and polymers)

– Non-local eﬀects (micro- and nano-devices)

– Non-linear eﬀects (high powers)

The main innovations of the theory are:

• To raise the dissipative ﬂuxes to the status of state variables

• To assign a central role to a generalized entropy, assumed to be a given

function of the whole set of variables, and whose rate of production is

always positive deﬁnite

Extended irreversible thermodynamics provides a decisive step towards a

general theory of non-equilibrium processes by proposing a unique formu-

lation of seemingly such diﬀerent systems as dilute and real gases, liquids,

polymers, microelectronic devices, nano-systems, etc. EIT is particularly well

suited to describe processes characterized by situations where the product of

relaxation time and the rate of variation of the ﬂuxes is important, or when

208 7 Extended Irreversible Thermodynamics

Table 7.2 Examples of application of EIT

High-frequency phenomena Short-wavelength phenomena

Ultrasounds in gases Light scattering in gases

Light scattering in gases Neutron scattering in liquids

Neutron scattering in liquids Heat transport in nano-devices

Second sound in solids Ballistic phonon propagation

Heating of solids by laser pulses Phonon hydrodynamics

Nuclear collisions Submicronic electronic devices

Reaction–diﬀusion waves in ecosystems Shock waves

Fast moving interfaces

Long relaxation times Long correlation lengths

Polyatomic molecules Rareﬁed gases

Suspensions, polymer solutions Transport in harmonic chains

Diﬀusion in polymers Cosmological decoupling eras

Propagation of fast crystallization fronts Transport near critical points

Superﬂuids, superconductors

the mean free path multiplied by the gradient of the ﬂuxes is high; these situ-

ations may be found when either the relaxation times or the mean free paths

are long, or when the rates of change in time and space are high. Table 7.2

provides a list of situations where EIT has found speciﬁc applications.

It should nevertheless not be occulted that some problems remain still

open like:

1. Concerning the choice of state variables:

– Are the ﬂuxes the best variables? Should it not be more judicious to

select a combination of ﬂuxes or a mixing of ﬂuxes and transport coef-

ﬁcients?

– Where to stop when the ﬂux of the ﬂux and higher-order ﬂuxes are taken

as variables? The answer depends on the timescale you are working

on. Shorter is the timescale, larger is the number of variables that are

needed.

– How far is far from equilibrium? In that respect, it should be convenient

to introduce small parameters related for instance to Deborah’s and

Knudsen’s numbers, allowing us to stop the expansions at a ﬁxed degree

of accurateness.

2. What is the real status of entropy, temperature, and the second law far

from equilibrium?

3. Most of the applications concern ﬂuid mechanics, therefore a description

of solid materials including polycrystals, plasticity, and viscoplasticity is

highly desirable.

4. The introduction of new variables increases the order of the basic diﬀeren-

tial ﬁeld equations requiring the formulation of extra initial and boundary

conditions.

5. Turbulence remains a challenging problem.

7.6 Problems 209

It should also be fair to stress that, during the last decade, many eﬀorts have

been spent to bring a partial answer to these acute questions.

7.6 Problems

7.1. Extended state space. Assume that the entropy s is a function of a vari-

able α and its time derivative η = dα/dt, and that α satisﬁes the diﬀerential

equation

M

d

2

α

dt

2

+

dα

dt

= L

∂s

∂α

.

(a) Show that the positiveness of the entropy production demands that L > 0

but does not imply any restriction on the sign of M. Hint: Write dα/dt and

dη/dt in terms of ∂s/∂α and ∂s/∂η. (b) Assume that ∂s/∂η = aη, with

a being a constant. Show that the stability condition d

2

s < 0 implies that

M > 0 and a = −(M/L).

7.2. Phase velocity. Determine the expressions (7.8) of the phase velocity v

p

and the attenuation factor α. Hint: Substitute the solution T(x, t) =

T

0

exp[i(kx−ωt)] writtenintheformT(x, t) = T

0

exp[i Rek(x−v

p

t)] exp(−x/α)

in the hyperbolic equation (7.5) and split the result in real and imaginary parts.

7.3. Non-local transport. (a) Check that the entropy and entropy ﬂux of the

non-local formalism including the ﬂux of the heat ﬂux presented in Sect. 7.1.3

are given by

ds =

1

T

du −

τ

1

v

λT

2

q dq −

τ

2

v

2λT

2

β

0

Q : d

0

Q−

τ

0

v

λT

2

β

QdQ

and

J

s

=

1

T

q +

1

λT

2

0

Q q +

1

λT

2

Qq.

(b) Prove that the entropy production corresponding to the Guyer–Krumhansl

equation (7.30) is

Tσ

s

=

3

τ

R

ρc

v

c

2

0

q q +

3τ

N

5ρc

v

_

(∇q) : (∇q)

T

+ 2(∇ q)(∇ q)

¸

.

(c) Show that the stationary heat ﬂux that satisﬁes the Guyer–Krumhansl

equation is the necessary condition for the entropy production to be a mini-

mum, under the constraint ∇ q = 0. In other terms, show that the Euler–

Lagrange equations corresponding to the variational equation

δ

_

(Tσ

s

−γ∇ q)dV = 0,

with respect to variations of q and γ are the steady state equations ∂u/∂t = 0

and ∂q/∂t = 0 provided one identiﬁes the Lagrange multiplier γ with twice

the temperature (Lebon and Dauby 1990).

210 7 Extended Irreversible Thermodynamics

7.4. Non-local transport. Show that (7.26)–(7.28) can be obtained by writing

for q and Q general evolution equations of the form

˙ q = −∇ Π+σ

q

,

˙

Q = ∇ Ξ+ σ

Q

,

at the condition that the ﬂux and source terms are given by the following

constitutive equations:

Π = A(T)I +QI +

0

Q, σ

q

= −q

and

Ξ = qI, σ

Q

= −Q,

wherein Ξ is a third-order tensor.

7.5. High-frequency wave speeds. (a) Obtain (7.32) from the energy balance

equation (7.2) and Guyer–Krumhansl relation (7.29), for longitudinal thermal

waves. (b) Verify that the high-frequency wave speed for longitudinal waves

is given by (7.33).

7.6. Continued-fraction expansions and generalized thermal conductivity. To

clarify the way to obtain the asymptotic expressions used in (7.36) and (7.40)

for the generalized thermal conductivity, (a) show that the continued fraction

R =

a

1 +

a

1+

a

1+...

tends, in the asymptotic limit of an inﬁnite expansion, to

R

∞

=

1

2

(

√

1 + 4a −1).

Hint: Note that, in this limit, R

∞

must satisfy R

∞

= a/(1 +R

∞

). (b) From

this result, and assuming that all correlation lengths are equal, check that

(7.36) leads to (7.39) for steady state (namely ω = 0) and for k = 2π/L.

7.7. Phonon hydrodynamics. Poiseuille ﬂow of phonons may be observed in

cylindrical heat conductors of radius R when the mean free paths

N

= c

0

τ

N

and

R

= c

0

τ

R

satisfy

N

R

¸ R

2

and

N

¸ R. In this case, (7.30) reduces

in a steady state to

∇T −

9

5

τ

N

ρc

v

∇

2

q = 0.

(a) Show that the total ﬂux along the cylinder will be given by

Q = −

5π

24

ρc

v

τ

N

R

4

∇T.

7.6 Problems 211

(b) Compare the dependence of Q with respect to the radius R with the

corresponding expression obtained from Fourier’s law. Note that this

dependence may be useful to describe the decrease of the eﬀective ther-

mal conductivity in very thin nanowires, in comparison with the usual

thermal conductivity of the corresponding bulk material.

7.8. Double time-lag behaviour. Instead of the Cattaneo’s equation (7.4),

some authors (Tzou 1997) use a generalized transport equation with two

relaxation times

τ

1

˙ q +q = −λ

_

∇T +τ

2

∇

˙

T

_

.

(a) Introduce this equation into the energy balance equation (7.2) and obtain

the evolution equation for the temperature. (b) Discuss the limiting behaviour

of high-frequency thermal waves.

7.9. Two-temperature models. Many systems consist of several subsystems,

each with its own temperature, as for instance, the electrons and the lattice

in a metal. It has been shown that the evolution equations for the electron

and lattice temperatures T

e

and T

l

are given, respectively, by:

c

e

∂T

e

∂t

= ∇ (λ∇T

e

) −C(T

e

−T

l

),

c

l

∂T

l

∂t

= C(T

e

−T

l

).

The constant C describes the electron–phonon coupling, which accounts for

the energy transfer from the electrons to the lattice, and c

e

and c

l

are the

speciﬁc heats of the electrons and lattice per unit volume, respectively. (a)

When the solution of the ﬁrst equation, namely T

e

= T

l

+ (c

l

/C)∂T

l

/∂t, is

introduced into the second one, prove that it leads to

∇

2

T

l

+

c

l

C

∂∇

2

T

l

∂t

=

c

l

+c

e

λ

∂T

l

∂t

+

c

e

c

l

λC

∂

2

T

l

∂t

2

.

(b) Show that this equation can also be obtained by eliminating q between the

energy balance equation and Guyer–Krumhansl’s equation, with the suitable

identiﬁcations of the parameters.

7.10. Limits of stability of non-equilibrium steady states. Let us ﬁrst consider

heat conduction in a rigid solid for which s = s(u, q). (a) Write the second

diﬀerential δ

2

s of the generalized entropy (7.25) around a non-equilibrium

state with non-vanishing value of q. (b) Show that one of the conditions to

be satisﬁed in order that the matrix of the second diﬀerential be negative

deﬁnite is

α

c

v

T

2

+

_

α

2

∂

2

α

∂u

2

−

_

∂α

∂u

_

2

_

q

2

≥ 0,

where α ≡ vτ/λT

2

. (b) If v, τ

1

, c

v

, and λ are constant, it is found that

∂α/∂u = −2α/(c

v

T) and ∂

2

α/∂u

2

= 6α(c

v

T)

−2

. Show that in this case the

former inequality is satisﬁed for values of q such that

212 7 Extended Irreversible Thermodynamics

q ≤

_

c

v

α

_

1/2

= ρc

v

T

_

λ

ρc

v

τ

1

_

1/2

,

where (λ/ρc

v

τ

1

)

1/2

= U is the second sound.

7.11. Development in gradients or in ﬂuxes. Compare the behaviour of the

wave number-dependent viscosity η(k, ω) appearing in the two following

second-order expansions (a) the ﬂux expansion

τ

2

∂P

v

∂t

+P

v

= −2ηV+∇

0

J

v

and

0

J

v

= −η

¸∇

0

P

v

),

with

0

J

v

the ﬂux of the viscous pressure tensor, η

a phenomenological co-

eﬃcient, and ¸. . .) the completely symmetrized part of the corresponding

third-order tensor and (b) the velocity gradient expansion

τ

2

∂P

v

∂t

+P

v

= −2ηV+

2

∇

2

V.

Note that (b) yields an unstable behaviour for high values of k, since the

generalized viscosity becomes negative. (For a discussion of these instabili-

ties arising in kinetic theory approaches to generalized hydrodynamics, see

Gorban et al. 2004.)

7.12. Two-layer model and the telegrapher’s equation. The so-called two-layer

model consists of a system whose particles jump at random between two

states, 1 and 2, with associated velocities v

1

= v and v

2

= −v, respectively,

along the x-axis. Assume that the rate R of particle exchange between the

two states per unit time and length is proportional to the diﬀerence of the

probability densities P

1

and P

2

, i.e. R = r(P

1

−P

2

). Show that the evolution

equations for the total probability density P = P

1

+P

2

and for the probability

ﬂux J = (P

1

−P

2

)v are, respectively,

∂P

∂t

+

∂J

∂x

= 0,

τ

∂J

∂t

+J = −D

∂P

∂x

,

with τ = 1/2r and D = v

2

τ (see van den Broeck 1990; Camacho and Zakari

1994).

7.13. Electrical system with resistance R and inductance L. The expression

relating the intensity I of the electrical current to the electromotive force ξ

is a relaxational equation given by

ξ = IR +L

dI

dt

,

7.6 Problems 213

with relaxation time τ

e

= L/R. The intensity I is related to the ﬂux of electric

current i by I = iA, with A the cross section of the conductor. The magnetic

energy stored in the inductor is given by U

m

=

1

2

LI

2

. Consider the total

internal energy U

tot

= U + U

m

, with U the internal energy of the material.

Show that the Gibbs’ equation dS = T

−1

dU +T

−1

p dV may be rewritten as

dS = T

−1

dU +T

−1

p dV −

τ

e

V

σ

e

T

i di,

which is a relation reminiscent of the Gibbs’ equation proposed in EIT. Hint:

Recall that R = (σ

e

A)

−1

l, with l the length of the circuit.

7.14. Chemical potential. According to (7.66), the diﬀerential equation for

the Gibbs’ free energy at constant temperature and pressure reads as

dg =

k

µ

k

dc

k

+

vτ

2η

P

v

: dP

v

,

and, after integration, g(c

i

, P

v

) = g

eq

(c

i

) +

1

4

vJP

v

: P

v

, where τ designates

the relaxation time of the viscous pressure tensor and J = η/τ, is the so-

called steady state compliance, a function which is often studied in polymer

solutions. (a) Obtain the modiﬁcation to the chemical potential of the ith

component, deﬁned by

µ

i

=

_

∂g

∂n

i

_

T,p,P

v

,

if the steady state compliance J is assumed to have the formJ = αM(ck

B

T)

−1

,

with M the molecular mass, c the number of macromolecules per unit vol-

ume, and α a constant (e.g. α = 0.400 for Rouse model of bead-and-spring

macromolecules, and α = 0.206 for the Zimm model, the latter including the

hydrodynamic interaction amongst the diﬀerent beads of a macromolecule

(Doi and Edwards, 1986; Bird et al., vol. 2, 1987a, 1987b)). (b) Study the in-

ﬂuence of the non-equilibrium contribution on the stability of the system; in

particular, determine whether the presence of a non-vanishing viscous pres-

sure will reinforce or not the stability with respect to that at equilibrium

(Jou et al. 2000).

Chapter 8

Theories with Internal Variables

The Inﬂuence of Internal Structure

on Dynamics

An alternative approach of non-equilibrium thermodynamics may be carried

out through the internal variables theory (IVT). This phenomenological de-

scription has been very successful in the study of a wide variety of processes in

rheology, deformable bodies, physico-chemistry and electromagnetism. IVT

may be considered as a generalization of the classical theory of irreversible

processes, just like extended irreversible thermodynamics. In the classical

theory, the state of the system is described by the local equilibrium vari-

ables, say energy, volume or deformation, but it is well known that many

non-equilibrium eﬀects cannot be adequately described by this set of vari-

ables. To avoid, among others, the inconvenience of a heavy formalism in

terms of functionals, one has, in extended irreversible thermodynamics, sup-

plemented the local equilibrium variables by the corresponding ﬂuxes. In the

IVT, one adopts a similar attitude by adding a certain number of internal

variables to the local equilibrium variables. Such variables are called internal

because they are connected either to internal motions or to local microstruc-

tures. They should not be confused with order parameters as introduced in

Chap. 6, which are describing some broken symmetry related to long-range

interactions as in phase transitions. We should also add that unlike the ﬂuxes

in extended irreversible thermodynamics, the internal variables are arbitrary

extensive quantities, not identiﬁed from the outset. It is the freedom in the

choice of the internal variables that explains the wide domain of applicability

of the formalism.

What is really meant by internal variables? The usual interpretation is that

they are macroscopic representations of some microscopic internal structures;

they can be of geometrical or physico-chemical nature, but generally there is

a priori no reference to their origin. The hope is that at the end, some phys-

ical meaning will come out from the theory itself. Mandel (1978) wrote that

“a clever physicist will always manage to detect internal variables and mea-

sure them”. Internal variables are essentially introduced to compensate for

our lack of knowledge about the behaviour of the system. The alternative

name of hidden variables is also found in the literature, as they refer to the

215

216 8 Theories with Internal Variables

internal structure that is hidden from the eyes of an external observer. At the

diﬀerence of the external variables, like temperature or volume, the internal

parameters are not coupled to external forces which provide a means of con-

trol. As they cannot be adjusted to a prescribed value through surface and

body forces, internal variables are not controllable. Being not attached to an

external force, they cannot take part in a mechanical work and therefore they

do not appear explicitly in the balance equations of mass, momentum and

energy. Additional equations (taking generally the form of rate equations)

are needed to describe their evolution.

To summarize, internal variables are in principle measurable but not con-

trollable and this makes diﬃcult their selection and deﬁnition. They are asso-

ciated to processes that are completely dissipated inside the system instead

of being developed against external surface or body forces.

Examples of internal variables are the extent of advancement in a chemical

reaction, the density of dislocations in plastic metals, mobile atoms in crys-

tals, the end-to-end distance between macromolecules in polymer solutions,

the shape and orientations of microstructures in macromolecules, the mag-

netization in ferromagnets, etc. In the following, we shall present the general

scheme of the thermodynamic theory of internal variables and afterwards we

shall discuss some applications both in ﬂuid and solid mechanics.

Following our policy of seeking a presentation as simple as possible, we

shall restrict our analysis to rather simple examples. The reader interested

by more complete developments is advised to consult the comprehensive re-

view papers by Bampi and Morro (1984), Kestin (1990, 1992), Maugin and

Muschik (1994), Lhuillier et al. (2003), and the well-documented book by

Maugin (1999) which includes a large bibliography and a wide and system-

atic presentation.

8.1 General Scheme

8.1.1 Accompanying State Axiom

The hypothesis of local equilibrium that is the basis of classical irreversible

thermodynamics is replaced in the IVT by the accompanying state axiom.

The essence of this postulate is that, to each non-equilibrium state, corre-

sponds an accompanying equilibrium state, and to every irreversible process is

associated an accompanying “reversible” process. This notion can be traced

back to Meixner (1973a, 1973b), Muschik (1990) and mainly to Kestin (1990,

1992): it provides an interesting picture of IVT by associating non-equilibrium

states with equilibrium states by means of a projection. The accompanying

equilibrium state is a ﬁctitious one; its state space 1 consists of the union of

the spaces of standard equilibrium variables ( and internal variables 1. The

8.1 General Scheme 217

ensemble 1 chosen for further analysis will consist of the following extensive

variables

1 := u, a, ξ; (8.1)

the space of classical variables includes the speciﬁc internal energy u and

the other local equilibrium variables, for instance the speciﬁc volume v in

ﬂuid mechanics or the elastic strain tensor ε in solid mechanics, which will

be generically designed as a, and ﬁnally ξ ≡ ξ

1

, ξ

2

, . . . , ξ

n

stands for a ﬁnite

set of internal variables (either scalars, vectors, or tensors). Although the

number of variables needed to specify a non-equilibrium state n is generally

much larger than the set (8.1), we can always imagine an accompanying

equilibrium state e whose extensive variables u, a, ξ take the same values as

in a non-equilibrium state n, i.e. u(e) = u(n), a(e) = a(n), ξ(e) = ξ(n). A

way of obtaining these accompanying state variables from the actual physical

non-equilibrium state is illustrated by the Gedanken experience of Fig. 8.1.

Consider a volume element n suddenly surrounded by an adiabatic rigid

enclosure so that no heat ﬂows and no external work can be performed;

as a consequence of the ﬁrst law,

˙

Q =

˙

W = 0 implies that u(n) = u(e).

Simultaneously, we constrain the internal variables ξ. Since no work and no

heat are exchanged, the system relaxes towards an equilibrium-constrained

state e with the values of u, a, and ξ unmodiﬁed. Observe however that

other variables, like the temperature T or the entropy s, relax to values

which are diﬀerent from their actual non-equilibrium values: T(e) ,= T(n) and

s(e) ,= s(n). The entropy s(e) is larger than the entropy in non-equilibrium

s(n) because the former is obtained from the latter by an adiabatic no-work

process. At this point it is interesting to emphasize the diﬀerence between the

axiom of accompanying state and the local equilibrium hypothesis. The gist

of this latter statement is to assume that the non-equilibrium entropy s(n)

and temperature T(n) are the same as in their accompanying equilibrium

state s(e) and T(e), respectively, while in IVT, the entropy s(e) is assumed

to characterize the equilibrium state.

Fig. 8.1 Producing an accompanying equilibrium e state from a non-equilibrium

state n

218 8 Theories with Internal Variables

Fig. 8.2 Accompanying reversible process

To clarify the notion of accompanying reversible process, consider an irre-

versible process between states n

1

and n

2

. The accompanying states e

1

and e

2

are obtained by projection on the state space (u, a, ξ) as shown in Fig. 8.2 and

the sequence of accompanying states between e

1

and e

2

constitutes the “re-

versible” accompanying process. By contrast with a true reversible process,

which takes place inﬁnitely slowly, the former one occurs at ﬁnite velocity

and, as seen later, is characterized by a non-zero entropy production.

Now the question remains of how many internal variables are needed for a

valuable description. Of course it is desirable to have a number of variables as

small as possible. There exists a criterion based on the notion of characteristic

time that allows deciding whether an internal variable is relevant or not. Let

us denote by τ

ξ

=

˙

ξ/ξ the intrinsic time of the internal mechanism and by

τ

a

= ˙ a/a the characteristic macroscopic time associated with the external

variable a; as usual, a upper dot denotes time derivative. If the Deborah

number deﬁned by De = τ

ξ

/τ

a

is such that De ¸1, then the corresponding

internal variable relaxes quickly towards its equilibrium value; this amounts

to say that the internal constraint is more or less removed, thus allowing

the system to reach a state of so-called unconstrained equilibrium (Kestin

1990, 1992). In contrast for De ¸1, the internal variable will remain frozen

during the timescale of the evolution of the variable a, we shall then speak

about a state of constrained equilibrium. Practically, in IVT only the internal

variables for which De ¸1 will be retained. It was also suggested by Kestin

(1990, 1992) to use the Deborah number as a measure of the “distance”

between a non-equilibrium state and an accompanying equilibrium state: the

smaller is De, the shorter is the distance between such states. Situations with

nonlocal eﬀects are discussed in Box 8.1.

8.1 General Scheme 219

Box 8.1 Non-Local Eﬀects. Histories of Variables

In presence of non-local eﬀects, as in the cases of damages or non-

homogeneous electrical or magnetic ﬁelds, it may in principle be necessary

to introduce spatial gradients ∇ξ of the internal variables in the description.

The evolution of such internal variables is now determined by complete bal-

ance equations, involving both a rate and a divergence term. The diﬃculty

is that such balance laws require appropriate boundary conditions, which

cannot in principle be assigned, as the internal variables are not control-

lable. This problem has not received suﬃcient attention and only partial

answers can be found in the recent literature. The underlying idea is that,

although the internal variables are not controllable, the second law of ther-

modynamics forces them to assume given values on the boundaries (Woods

1975; Valanis 1996; Cimmelli 2002).

Because of the diﬃculty to identify the nature of the internal variables,

it may be tempting to formulate the problem without having recourse to

them. To obtain such a description, it suﬃces to eliminate the internal

variables ξ from (8.11) to (8.13). In the case that (8.13) can be solved,

we can substitute its solutions in (8.11) and (8.12), leading to a set of

constitutive relations, which turn out to be functionals Q and T of the

histories of u (or T), ∇T, and a (Kestin 1992):

q = Q

t

≥0

¦T(t −t

), ∇T(t −t

), a(t −t

)¦, (8.1.1)

F = T

t

≥0

¦T(t −t

), ∇T(t −t

), a(t −t

)¦. (8.1.2)

As shown in Chap. 9, it is customary in rational thermodynamics to claim

that the systems that it describe are materials with memory. The diﬃculty

is how to measure and to catalogue all these histories and how to solve the

subsequent integro-diﬀerential equations obtained after substitution of the

above functionals in the balance equations.

8.1.2 Entropy and Entropy Production

The axiom of accompanying equilibrium state leads us to accept the existence

of an accompanying entropy s that is a function of the whole set 1 of variables:

s = s(u, a, ξ). (8.3)

The corresponding Gibbs’ equation in rate form will assume the form

˙ s = T

−1

˙ u −T

−1

ρ

−1

F

e

˙ a +T

−1

A

˙

ξ, (8.4)

where T is the temperature, F

e

the mechanical force conjugate to the ob-

servable a, and A the aﬃnity (also known as conﬁgurational or Eshelby

220 8 Theories with Internal Variables

force in some branches of solid mechanics) conjugate to ξ. They are deﬁned,

respectively, by

1

T

=

∂s

∂u

,

F

e

ρT

= −

∂s

∂a

,

A

T

=

∂s

∂ξ

. (8.5)

The internal energy obeys the balance law

ρ ˙ u = −∇ q +F ˙ a (8.6)

with q the heat ﬂux vector. Remark in passing that the forces F(n) required

to produce work on the volume element in the actual physical space are diﬀer-

ent from the forces F

e

in (8.4) which represent these acting on the ﬁctitious

reversible process. After elimination of ˙ u between the Gibbs’ equation (8.4)

and the energy balance equation (8.6), one obtains the expression for the

evolution of the entropy of the accompanying state during the process:

ρ˙ s +∇ (qT

−1

) = q ∇T

−1

+T

−1

(F −F

e

) ˙ a +ρT

−1

A

˙

ξ. (8.7)

Suppose that the entropy obeys a balance equation of the standard form

ρ˙ s +∇ (T

−1

q) = σ

s

≥ 0, (8.8)

wherein it is accepted that the rate of entropy production σ

s

is positive

deﬁnite and the entropy ﬂux given by T

−1

q. By comparison with (8.7), it is

then found that

σ

s

= q ∇T

−1

+T

−1

(F −F

e

) ˙ a +ρT

−1

A

˙

ξ ≥ 0. (8.9)

With the exception of extended thermodynamics, it is currently admitted

that, in absence of diﬀusion of matter, the entropy ﬂux is the ratio of the heat

ﬂux and the temperature. The positiveness of the rate of entropy production is

a more disputed question and can be considered as a supplementary postulate

of the theory.

Instead of starting from (8.9), some authors prefer to work with the

Clausius–Duhem’s inequality

−ρ(

˙

f +s

˙

T) +F ˙ a +T

−1

q ∇T ≥ 0, (8.10)

which is directly obtained from (8.8) when use is made of the deﬁnition of

the free energy f = u − Ts and the energy balance equation (8.6). In the

future, we shall indiﬀerently start from (8.9) or (8.10).

8.1.3 Rate Equations

It is noted that (8.9) has the form of the familiar sum of generalized forces

X and ﬂuxes J, which, at the exception of the heat ﬂux, are the rate of

change of the extensive variables. In classical irreversible thermodynamics

8.2 Applications 221

(see Chap. 2), the ﬂuxes are linear expressions of the forces but this is not

mandatory in the present theory wherein the ﬂuxes and forces are related

to each other by general functions containing the variables u, a, and ξ, i.e.

J = J(X; u, a, ξ). Such ﬂux–force relations take generally the form of rate

equations, which conventionally will be written as

q = Q(∇T, ˙ a, A; u, a, ξ), (8.11)

F −F

e

= T(∇T, ˙ a, A; u, a, ξ), (8.12)

˙

ξ = Ξ(∇T, ˙ a, A; u, a, ξ), (8.13)

where Q, T, and Ξ denote functions of their arguments.

After this short digression, let us go back to the evolution equation (8.13).

In the simplest case, we may assume that the “ﬂux”

˙

ξ is linear with the

“force” A

˙

ξ = lA(u, a, ξ), (8.13)

where l is a phenomenological coeﬃcient. Such a simpliﬁed evolution equation

is justiﬁed when the thermal gradients play a negligible role and when F ≈ F

e

as often occurring in solid mechanics. The entropy production (8.9) takes now

the simple form

Tσ

s

= ρA

˙

ξ, (8.14)

and we shall examine three particular cases:

1. A =

˙

ξ = 0: the system has reached a state of unconstrained equilibrium

characterized by A(u, a, ξ) = 0 or ξ(e) = ξ(u, a).

2. A ,= 0 with

˙

ξ = 0: this is a case of constrained equilibrium still deﬁned

by a zero entropy production.

3. A ,= 0 and

˙

ξ ,= 0: the system relaxes towards the equilibrium surface

ξ(e).

As mentioned earlier, there exists a wealth of applications of the inter-

nal variable theory, mainly in the domain of material sciences. Here we will

present only three of them, and we refer the reader to specialized books and

articles for a more exhaustive list of examples.

8.2 Applications

8.2.1 Viscoelastic Solids

Viscoelastic solids are a privileged ﬁeld of application of the theory of in-

ternal variables. Thermodynamics of viscoelasticity in solids was initiated

by Bridgman (1941) and Eckart (1948) and developed later on by Eringen

(1967), Kluitenberg (1962) and Kestin and Bataille (1980), among others.

Our purpose is to discuss qualitatively some essential features of rheological

222 8 Theories with Internal Variables

models like those of Maxwell and Kelvin–Voigt, which have played a central

role in the theory of viscoelasticity; excessive generality will be deliberately

avoided.

We restrict our attention to isothermal eﬀects and make the assumption

that the solid is a one-dimensional homogeneous and isotropic rod subject

to inﬁnitesimally small deformations. Under the action of a stress σ the solid

undergoes a strain ε. It is remembered from Sect. 8.1 that an intensive para-

meter such as σ, which is the stress actually applied in the non-equilibrium

state, is not the same as the stress σ

e

in the accompanying equilibrium state.

Moreover, it is assumed that the deformation of the body has, associated

with it, an internal work that is completely dissipated in the interior of the

system during the actual dissipative process. Its expression is given by /dξ

where / is the “aﬃnity” associated to the internal variable ξ. According to

IVT, the space of basic variables is constituted by 1 ≡ u, σ

e

, ξ with u the

speciﬁc internal energy. The entropy associated with the accompanying state

satisﬁes the Gibbs’ equation

T ds = du −ρ

−1

σ

e

dε +/dξ, (8.15)

where ρ is the constant mass density. Since the temperature is a constant, it

may be preferable to work with the free energy f for which

df = ρ

−1

σ

e

dε −/dξ, (8.16)

with

ρ

−1

σ

e

= ∂f/∂ε, / = −∂f/∂ξ. (8.17)

The equation of state f = f(ε, ξ) can be expanded in a Taylor series around

a reference state ε = ξ = 0 and put in the form

f(ε, ξ) =

1

2

Eε

2

+Bεξ +

1

2

Cξ

2

, (8.18)

where E, B, and C are constant coeﬃcients with E > 0, C > 0, and B

2

≤ EC

to satisfy the stability property of the equilibrium state. Examples of linear

state equations (8.17) are

ρ

−1

σ

e

= Eε +Bξ, (8.19)

−/ = Bε +Cξ. (8.20)

Elimination of ˙ u from the energy balance equation

ρ ˙ u = σ : ˙ ε, (8.21)

and the Gibbs’ equation (8.15) written in rate form leads to the following

relation for the evolution of the entropy in the accompanying state:

ρT ˙ s = (σ −σ

e

) : ˙ ε +ρ

−1

A

˙

ξ. (8.22)

8.2 Applications 223

Since there is no heat ﬂux, the entropy ﬂux will be zero and therefore the

quantity ρT ˙ s can be identiﬁed with the rate of dissipated energy Tσ

s

, with

σ

s

the rate of entropy production. In very slow processes, it is justiﬁed to

assume (Kestin and Bataille 1980) that σ = σ

e

so that the dissipated energy

acquires the simple form

ρTσ

s

= A

˙

ξ ≥ 0. (8.23)

Of course, the simplest rate equation for the internal variable is to put

˙

ξ = lA(l > 0). (8.24)

Taking the time derivative of (8.19) and using (8.24), one obtains

ρ

−1

˙ σ = E ˙ ε +lB/, (8.25)

where use is made of the assumption σ

e

= σ. Substituting now / from (8.20)

and eliminating ξ from (8.19), leads to the following relaxation equation for

the stress tensor

τ

ε

˙ σ +σ = E

∞

(ε +τ

σ

˙ ε), (8.26)

where τ

ε

= 1/lC, τ

σ

= E(lCE

∞

)

−1

, E

∞

= E −B

2

C

−1

. It is worth noticing

that relation (8.26) is the constitutive equation for a Poynting–Thomson

body. By letting τ

ε

and τ

σ

go to zero, one ﬁnds back the Hooke’s law of

elasticity, and E

∞

is the classical Young modulus. Setting τ

ε

= 0, one recovers

the Kelvin–Voigt model

σ = E

∞

ε +η ˙ ε, (8.27)

with η = E/lC, while by putting E

∞

= 0 and E

∞

τ

σ

= η in (8.26), one

obtains Maxwell’s model:

τ

ε

˙ σ +σ = η ˙ ε. (8.28)

It is therefore claimed that the thermodynamics of internal variables includes

the formulation of the standard models of rheology. The above considerations

have been generalized to ﬁnite strains by Sidoroﬀ (1975, 1976), among others.

IVT has been used with success in several branches of solid mechanics. For

instance, an interesting approach of plasticity, in relation with the microscopic

theory of dislocations, was developed by Ponter et al. (1978). These authors

show that the area swept out by the dislocation lines is a meaningful internal

variable. More explicitly, the internal variable selected by Ponter et al. is

ξ =

ˆ

ANb

3

, (8.29)

where

ˆ

A is the area swept out by the dislocation line, N the number of non-

interacting so-called Franck–Read’s sources, b the Burgers vector associated

with dislocations slip, and

3

the corresponding volume. Another ﬁeld of

application is the theory of fracture, where the internal variable can be chosen

as a scalar variable D (D for damage) such that 0 ≤ D ≤ 1, D = 0 meaning

absence of cracks and D = 1 corresponding to fracture.

224 8 Theories with Internal Variables

8.2.2 Polymeric Fluids

As a second illustration, we shall examine the behaviour of polymeric ﬂuids,

which are one of the great diversity of ﬂuids called complex ﬂuids. Their

characteristic property is to possess a certain degree of internal structures

or microstructures. Therefore they behave very diﬀerently from water or air

when forces are exerted on them. The most rigorous method of describing

materials with internal microstructures is through the distribution function

ψ which depends on the position and velocity vectors of the ensemble of mass

points. However, in view of the large number of variables, this approach be-

comes quickly impracticable, and some mechanical approximations taking the

form of rigid rods, dumbbells or more complicated assemblages (see Fig. 8.3)

have been introduced.

In the case of dumbbells, the description involves a seven-dimensional func-

tion ψ(r, R, t) where r is the position vector and R the end-to-end macro-

molecular distance. But still more approximations are needed to simplify the

model. In that respect, it is frequent to replace the description based on the

distribution function by the second-order symmetric tensor

C =

_

RRψ dR, (8.30)

called “conﬁguration” or “conformation” tensor. The advantage gained by

this approximation is to solve model equations in a large variety of ﬂow sit-

uations. It is the conformation tensor that we shall identify as the relevant

internal variable for our forthcoming analysis. We consider a ﬂowing incom-

pressible ﬂuid composed of a carrier, the solvent of mass density ρ

f

, and long

macromolecular chains, the solute, of mass density ρ

pol

. Because of incom-

pressibility, the total mass ρ = ρ

f

+ρ

pol

is a constant and ∇ v = 0, where v

designates the barycentric velocity deﬁned by means of ρv = ρ

f

v

f

+ρ

pol

v

pol

.

The selected basic variables are the mass concentration c

pol

= ρ

pol

/ρ of

the polymeric chain, the barycentric velocity v, the speciﬁc internal energy

u or the temperature T and the conformation tensor C. The corresponding

evolution equations are given by

Fig. 8.3 Molecular polymer models

8.2 Applications 225

ρ˙ c

pol

= −∇ J, (8.31)

ρ ˙ v = ∇ σ +ρF, (8.32)

ρ ˙ u = −∇ q + σ

v

: V, (8.33)

˙

C = ((C, V, c

pol

, u), (8.34)

wherein J denotes the ﬂux of matter, F the external body force per unit

mass, V the symmetric velocity gradient tensor, ( a unknown function of the

variables C, c

pol

, u and eventually of V; ﬁnally, σ is the symmetric stress

tensor and σ

v

(= σ + pI) its viscous part with p the hydrostatic pressure.

The tensor σ

v

will be inﬂuenced by the presence of the internal variable

and therefore will depart from the Newton law by an additional contribution

so that in total

σ

v

= 2ηV+ σ

pol

(C, V, . . .), (8.35)

where η is the usual shear viscosity coeﬃcient. The system of equations

(8.31)–(8.34) will be closed by providing constitutive relations for the set

¦J, q, σ

v

(or σ

pol

), C¦. These are obtained from the axiom of the local accom-

panying state consisting in formulating a generalized Gibbs’ equation and

calculating the corresponding rate of entropy production inside the material.

Let us write the Gibbs’ relation in terms of the free energy, i.e.

df = −s dT +µdc

pol

−A : dC, (8.36)

where µ = µ

pol

− µ

f

is the relative chemical potential, A is a symmetric

second-order tensor conjugate to the internal variable C describing the way

in which the free energy depends on the internal variable:

A = −

∂f

∂C

. (8.37)

There is no term of the form p d(ρ

−1

) in (8.36) because it is assumed that

the mass density is constant. By combining relation (8.36) with the balance

equations of mass concentration (8.31) and energy (8.33) and comparing with

the general evolution equation for the entropy

ρ˙ s = −∇ J

s

+σ

s

, (8.38)

one obtains the following results for the entropy ﬂux and the entropy pro-

duction, respectively:

J

s

=

1

T

(q −µJ), (8.39)

Tσ

s

= −J

s

∇T −J ∇µ + σ

v

: V+ρA :

˙

C ≥ 0. (8.40)

Expression (8.39) of the entropy ﬂux is classical while the dissipated energy

Tσ

s

contains a supplementary contribution compared to the similar relation

for a Newtonian ﬂuid. For the time being, limiting ourselves to expressions of

226 8 Theories with Internal Variables

the ﬂuxes J

s

, J, σ

v

, and

˙

C, which are linear and isotropic in the forces ∇T,

∇µ, V, and A, one is led to write

J

s

= −λ∇T −χ∇µ, (8.41)

J = −χ∇T −D∇µ, (8.42)

σ

v

= 2ηV+lA, (8.43)

˙

C = kA−lV = −k

∂f

∂C

−lV, (8.44)

wherein λ, χ, D, η, l, and k are phenomenological coeﬃcients with λ > 0,

η > 0, D > 0, and k > 0. The same coeﬃcient χ (respectively, l) appears

in (8.41) and (8.42) (respectively, in (8.43) and (8.44)) because of Onsager’s

reciprocal property, the change of sign in front of l in (8.43) and (8.44) is a

consequence of the property that the ﬂuxes σ

v

and

˙

C are of opposite parity

with respect to time reversal. The terms lA in (8.43) and −lV in (8.44) do

not contribute to the entropy production and have therefore been labelled as

gyroscopic or non-dissipative. In most complex ﬂuids, these non-dissipative

contributions may also contain non-linear terms (Lhuillier and Ouibrahim

1980) resulting from the coupling of A (or V) with C and (8.43) and (8.44)

will more generally be of the form:

σ

v

= 2ηV+lA−β(C A+A C), (8.45)

˙

C = k

∂f

∂C

−lV+β(C V+V C). (8.46)

The main results of the model are contained in (8.45) and (8.46) which show

that there is a strong interconnection between the way the molecular deforma-

tion C inﬂuences the stress tensor and the way the rate of ﬂuid deformation

modiﬁes the molecular deformation. A last remark is in form: if it is wished

to satisfy the axiom of material frame-indiﬀerence (see Chap. 9), the mater-

ial time derivative in (8.46), which is non-objective, must be replaced by an

objective one, as for instance the Jaumann time derivative

D

J

C =

˙

C+W C−C W, (8.47)

with W the skew-symmetric velocity gradient tensor. Moreover, there will

be no diﬃculty to generalize (8.45) and (8.46) with terms of third or higher

orders in C, V, and A, but we shall see that the present description oﬀers

still enough possibilities (see Box 8.2).

Box 8.2 Simpliﬁed Models

Several simpliﬁcations of the model (8.45) and (8.46) are of interest:

(1) If in relation (8.46) we set β = 1 and = 0, one recovers a model

obtained in the kinetic theory. In this case the additional stress deﬁned

by (8.35) is simply σ

pol

= 2A C, but the kinetic theory predicts

8.2 Applications 227

instead that σ

pol

≈ A from which we conclude that the results of the

kinetic theory are in contradiction with the IVT. The problem encoun-

tered in the kinetic theory ﬁnds probably its origin in the diﬃculties

arising in the evaluation of σ

pol

.

(2) Let us still assume that l = 0 but that the free energy f is quadratic

in C so that A = −∂f/∂C = −HC where H is some kind of Hookean

spring constant; (8.45) and (8.46) read then as

σ

v

= 2ηV+ 2βHC C, (8.2.1)

D

J

C = −

1

τ

C+β(C V+V C). (8.2.2)

This model was proved to be at least as valuable as the model of Oldroyd

(Bird et al. 1987) widely used in rheology; (8.2.1) exhibits clearly the

dependence of the non-Newtonian stress on the Hookean elasticity of

the polymer, whereas (8.2.2) is of the relaxation type with τ(= 1/kH), a

positive relaxation time. The physical interpretation of (8.2.2) is rather

evident: the presence of the Jaumann time derivative forces the mole-

cules to rotate with the surrounding ﬂuid, the rate of ﬂuid deformation

V deforms the molecules with an eﬃciency β and the molecules relax

towards an equilibrium conﬁgurational state with a relaxation time τ.

(3) Finally, by admitting that β = 0 and f quadratic in C, relation (8.46)

reduces to

D

J

C = −

1

τ

C−lV. (8.2.3)

Such an expression presents a striking resemblance with an equation

derived by Frankel and Acrivos (1970) to describe emulsions of quasi-

spherical drops with C a second-order tensor, expressing the deviation

of drops from sphericity.

8.2.3 Colloidal Suspensions

Colloidal suspensions are another example of complex ﬂuids. A colloidal sus-

pension is a collection of particles suspended in a ﬂuid, called the solvent or

the carrier. The driven particles must have a size larger than 10

−8

m for the

particles to be considered as a continuum, but smaller than 10

−4

m to avoid

sedimentation. They may be aerosols (fog, smog, smoke), emulsiﬁed drops,

ink, paint or asphalt particles. The suspending ﬂuid must be suﬃciently dense

in order that its mean free path be smaller than the size of the particles, thus

excluding rareﬁed gases. These hypotheses allow us to consider the system as

made by two interpenetrating continuous media. To simplify the issue, it is

generally admitted that the particles are rigid spheres of the same size mov-

ing in a Newtonian ﬂuid so that the system may be considered as isotropic.

228 8 Theories with Internal Variables

In the foregoing analysis, we shall omit the role of velocity ﬂuctuations whose

treatment can be found in specialized literature, e.g. Lhuillier (1995, 2001).

It is our aim to show that IVT provides a valuable tool for the study

of thermal diﬀusion in suspensions; the supramolecular size of the particles

being taken into account by the introduction of an internal variable. After

commenting in Sect. 8.2.3.1 the diﬀerences between particles suspensions and

molecular mixtures, we will present the evolution equations of the relevant

parameters and ﬁnally discuss the restrictions placed by the second law of

thermodynamics.

8.2.3.1 Suspensions vs. Molecular Mixtures

It is true that the suspensions of particles in a ﬂuid behave in several aspects

as a binary molecular mixture of solute molecules in a solvent. However,

the particle size is larger than the molecular dimensions and this has two

important consequences. First, the volume of a particle is well deﬁned whereas

the volume of a molecule is not. A description of suspensions will therefore

include not only the particle mass fraction c(= dM

p

/dM) but also its volume

fraction φ(= dV

p

/dV ) which is related to c by

φ = ρc/ρ

p

, (8.48)

where ρ is the total mass per unit volume of the suspension

ρ = φρ

p

+ (1 −φ)ρ

f

, (8.49)

the quantity ρ

p

(= dM

p

/dV

p

) is the mass per unit volume of the material from

which the particles are made and ρ

f

(= dM

f

/dV

f

) the mass per unit volume

of pure ﬂuid; from now on, subscripts p and f will refer systematically to

“particle” and “ﬂuid”, respectively. Moreover, to describe suspensions it is

necessary to introduce a volume-weighted velocity u deﬁned by

u = φv

p

+ (1 −φ)v

f

, (8.50)

besides the classical barycentric mass-weighted velocity v,

v = cv

p

+ (1 −c)v

f

. (8.51)

The second diﬀerence between suspensions and molecular mixtures is the

large inertia of particles from which follows that the average particle velocity

v

p

can be very diﬀerent from the ﬂuid velocity v

f

. Hence the importance of

the particle diﬀusion ﬂux

J = ρc(v

p

−v) = ρc(1 −c)(v

p

−v

f

), (8.52)

which is no longer given by the Fick’s law but will obey a full evolution

equation, as shown below. It is not diﬃcult to convince oneself of the relations

8.2 Applications 229

u = v + (ρ

−1

p

−ρ

−1

f

)J, (8.53)

φ = c + (1 −c)(ρ

−1

p

−ρ

−1

f

), (8.54)

showing that u and φ are generally diﬀerent from v and c respectively except

for the special cases that particles and ﬂuid have similar mass densities.

8.2.3.2 State Variables and Evolution Equations

In presence of thermal gradients, the system will exhibit thermodiﬀusion.

Since we are in presence of two phases, a rather natural choice of variables

will be mass densities of phases, the average temperature and the velocities of

the particles and ﬂuid phase. However, in analogy with molecular diﬀusion,

a more appropriate selection of variables is the following

ρ, c, T, v, ξ(= v

p

−v

f

), (8.55)

where the relative velocity will be considered as an internal variable. More

reﬁned descriptions will require the introduction of more internal variables

as for instance a structural vector R related to the statistical distribution of

the particles (Lhuillier 2001) but, for the present purpose, it is suﬃcient to

restrict our analysis to the set (8.55). The evolution equations for the classical

variables ρ, c, T, and v are given by

˙ ρ = −ρ∇ v (total mass balance), (8.56)

ρ˙ c = −∇ J (mass concentration balance), (8.57)

ρ ˙ v = −∇p +∇ Π

v

+ρg(momentumbalance), (8.58)

ρ ˙ u = −∇ q −p∇ v +Π

v

: (∇v)

s

(internal energy balance), (8.59)

where p is the hydrostatic pressure, Π

v

is the symmetric viscous stress tensor,

(∇v)

s

the symmetric part of the tensor ∇v, and u the internal energy per unit

mass. It remains to determine the evolution equation of the internal variable

ξ, i.e. the relative velocity v

p

− v

f

. The latter is obtained from the two

momentum equations for the particulate and ﬂuid phases given, respectively,

by Lhuillier (1995)

ρc

d

p

v

p

dt

= −φ(∇p −∇ σ

v

) +f +ρcg, (8.60)

ρ(1 −c)

d

f

v

f

dt

= −(1 −φ)(∇p −∇ σ

v

) −f −ρ(1 −c)g, (8.61)

where two new material time derivatives are introduced d

p

/dt = ∂/∂t +

v

p

∇ for the particles and a similar expression for the ﬂuid; f is the mutual

interaction force between the particles and the ﬂuid measured per unit volume

of the suspension. At this stage, f is an unknown quantity whose constitutive

230 8 Theories with Internal Variables

equation will be determined later on, σ

v

is the so-called suspension viscous

stress related to the stress Π

v

by Lhuillier (1995)

Π

v

= σ

v

−ξJ, (8.62)

besides the viscous term σ

v

, one observes the presence of a stress ξJ of kinetic

origin which is introduced to recover the global momentum equation (8.58).

Making use of the following identity

˙

ξ =

d

p

v

p

dt

−

d

f

v

f

dt

+∇

_

1

2

(2c −1)ξ

2

_

+ξ ∇v, (8.63)

one obtains the following evolution equation for the internal variable ξ:

˙

ξ = (ρ

−1

f

−ρ

−1

p

)(∇p−∇ σ

v

) +

f

ρc(1 −c)

+∇

_

1

2

(2c −1)ξ

2

_

−ξ ∇v. (8.64)

Set (8.56)–(8.59), (8.64) contains some indeterminate unknown quantities like

p, u, σ

v

, q, and f whose expressions will be determined from thermodynam-

ics and more particularly from the positive deﬁnite property of the entropy

production

σ

s

= ρ˙ s +∇ J

s

≥ 0. (8.65)

Of course at this stage of the discussion, the entropy s and the entropy ﬂux

J

s

remain undetermined quantities to be expressed by means of constitutive

relations. Let us now examine the consequences issued from the positiveness

of σ

s

.

8.2.3.3 Restrictions Placed by the Second Law

of Thermodynamics

The total kinetic energy per unit mass can be written as

1

2

cv

2

p

+

1

2

(1 −c)v

2

f

=

1

2

v

2

+

1

2

c(1 −c)ξ

2

. (8.66)

To keep the usual expression

1

2

v

2

for the kinetic energy per unit mass, we

shall admit that the part of the kinetic energy involving the relative velocity

pertains to the internal energy so that

u(s, ρ, c, ξ) = u

0

(s, ρ, c) +

1

2

c(1 −c)ξ

2

, (8.67)

where u

0

is the local equilibrium energy depending exclusively on the set of

“equilibrium” variables. The corresponding Gibbs’ equation, written in rate

form, will therefore be given by

˙ u = T ˙ s +

p

ρ

2

˙ ρ +µ˙ c +

J

ρ

˙

ξ. (8.68)

8.2 Applications 231

Note that the chemical potential is now including a kinetic contribution and

is related to the local equilibrium chemical potential µ

0

(= ∂u

0

/∂c) by

µ = µ

0

+

1

2

(1 −2c)ξ

2

. (8.69)

After elimination of ˙ u, ˙ ρ, ˙ c, and

˙

ξ by means of the evolution equations (8.56)–

(8.59) and (8.64), we obtain the following entropy balance

T(ρ˙ s +∇ J

s

) ≡ Tσ

s

= ∇ [TJ

s

−q −µJ + (u −v) σ

v

] + σ

v

: (∇u)

sym

−J

s

∇T −J

_

∇µ

0

−(ρ

−1

p

−ρ

−1

f

)∇p +f [ρc(1 −c)]

−1

_

≥ 0. (8.70)

Positiveness of the dissipated energy Tσ

s

requires that the divergence term

in (8.70) vanishes, whence the following expression for the entropy ﬂux:

J

s

=

1

T

(q −µJ) −

1

T

(u −v) σ

v

. (8.71)

The ﬁrst two terms in (8.71) are classical but a new term depending on the

relative velocity and the mechanical stress tensor is appearing. The remaining

terms in (8.70) take the form of bilinear products of thermodynamic ﬂuxes

and forces. The simplest way to guarantee the positive deﬁnite character of

the dissipated energy is to assume that these ﬂuxes and forces are related by

means of linear relations, i.e.

σ

v

= η(∇u)

s

, (8.72)

J

s

= −

λ

T

∇T + ˜ sJ, (8.73)

f = −ρc(1 −c)

_

˜ s∇T −(ρ

−1

p

−ρ

−1

f

)∇p +∇µ

0

+D

−1

J

¸

, (8.74)

the phenomenological coeﬃcients η, λ, ˜ s, D depend generally on ρ, c, and T,

the same ˜ s appears in both (8.73) and (8.74) to satisfy the Onsager reciprocal

relations. After substitution of the ﬂux–force relations (8.72)–(8.74) in (8.70)

of the dissipated energy, one is led to

Tσ

s

=

λ

T

(∇T)

2

+

1

D

J

2

+

1

η

σ

v

: σ

v

≥ 0, (8.75)

from which follows that λ > 0, η > 0, D > 0, there is no restriction on the

sign of ˜ s. The above results warrant further comments. It is important to note

that the stress σ

v

is related to the gradient of the volume-weighted velocity

u rather than to the gradient of the mass-weighted velocity v as in molecular

diﬀusion. This property has been corroborated by microscopic considerations

and is a well-known result in the theory of suspensions. Relation (8.73) can be

viewed as an expression of the Soret law stating that a temperature gradient

is capable of inducing a ﬂux of matter.

The result (8.74) is important as it provides an explicit relation for the

inter-phase force f between the particles and the ﬂuid, and plays, for sus-

pensions, the role of Fick’s law for binary mixtures. This interaction force

232 8 Theories with Internal Variables

will ultimately appear as a sum of elementary forces involving ∇c (through

(∂µ

0

/∂c)

p,T

) (concentration-diﬀusion force), ∇p (baro-diﬀusion force), ∇T

(thermodiﬀusion force) and the relative velocity v

p

− v

f

(through D

−1

J)

(kinematic-diﬀusion force).

Concerning the concentration-diﬀusion force, it always drives the particles

towards regions of lower particles concentration because (∂µ

0

/∂c)

p,T

> 0,

which is a consequence of thermodynamic stability. Experimental investiga-

tions conﬁrm that the concentration-diﬀusion force is the most important,

that the thermodiﬀusion force is rather small, and that the baro-diﬀusion

force is negligible.

When the two following conditions are satisﬁed, ρ

p

= ρ

f

, and d

p

v

p

/dt =

d

f

v

f

/dt, it is found by subtracting (8.60) from (8.61) that the force f van-

ishes identically. If in addition the temperature and pressure are kept con-

stant, (8.74) boils down to Fick’s law J = −D∇µ

0

, where D is the positive

coeﬃcient of diﬀusion.

Expression (8.74) of f is sometimes decomposed into a “non-dissipative”

force f

∗

(the three ﬁrst terms under brackets in (8.74)) and a dissipative

contribution, namely

f = f

∗

−γ(v

p

−v), (8.76)

where use is made of the deﬁnition (8.52) of J and where γ, called the friction

coeﬃcient, stands for γ = ρ

2

c

2

(1 − c)/D > 0. The term “non-dissipative” is

justiﬁed as it corresponds to situations for which γ = 0, i.e. D = ∞, which

is typical of absence of dissipation.

The expression of the heat ﬂux vector q is directly derived by eliminating

J

s

between (8.71) and (8.73); making use of (8.72) and introducing a pseudo-

enthalpy function

˜

h = T ˜ s +µ, it is found that

q = −λ∇T +

˜

hJ +η(u −v) ∇u. (8.77)

For pure heat conduction, one recovers the classical Fourier law so that the

coeﬃcient λ can be identiﬁed with the heat conductivity. For a molecular

mixture for which u = v, the above relation is equivalent to the law of

Dufour, expressing that heat can be generated by matter transport.

The above analysis shows that internal variables oﬀer a valuable approach

of the theory of suspensions. It is worth noticing that the totality of results

obtained in this section was also derived in the framework of extended irre-

versible thermodynamics (Lebon et al. 2007).

8.3 Final Comments and Comparison

with Other Theories

Thermodynamics with IVT provides a rather simple and powerful tool for

describing structured materials as polymers, suspensions, viscoelastic bodies,

electromagnetic materials, etc. As indicated before, its domain of applicability

8.3 Final Comments and Comparison with Other Theories 233

is very wide, ranging from solid mechanics, hydrodynamics, rheology, electro-

magnetism to physiology or econometrics sciences. IVT requires only a slight

modiﬁcation of the classical theory of irreversible processes by assuming that

the non-equilibrium state space is the union of two subsets. The ﬁrst one is

essentially composed by the same variables as in classical irreversible ther-

modynamics while the second subset is formed by a more or less large set

of internal variables that have two main characteristics: ﬁrst, they cannot be

controlled by an external observer and second, they can be unambiguously

measured. Furthermore, it is assumed that to any irreversible process, one

can associate a ﬁctitious reversible process referred to as the accompanying

state. It was also proved that by eliminating one or several internal variables,

one is led to generalized constitutive relations taking the form of functional

of the histories of the state variables. In that respect, it can be said that the

IVT is equivalent to rational thermodynamics (see next chapter).

The main diﬃculty with IVT is the selection of the number and the iden-

tiﬁcation of the nature of the internal variables. It is true that for some

systems, like polymers or suspensions, the physical meaning of these vari-

ables can be guessed from the onset, but this is generally not so. In most

cases, the physical nature of the internal variables is only unmasked at the

end of the procedure. In some problems, like in suspensions, the dependence of

the thermodynamic potentials on these extra variables is a little bit “forced”.

Referring for instance to (8.67) of the internal energy u, it is not fully justi-

ﬁed that the dependence of u on the internal variable ξ is simply the sum of

the local internal energy and the diﬀusive kinetic energy. Another problem is

related to the time evolution of the internal variables. Except some particu-

lar cases, like diﬀusion of suspensions, there is no general technique allowing

us to derive these evolution equations, in contrast with extended irreversible

thermodynamics or GENERIC (see Chap. 10). Moreover, as these variables

are in principle not controllable through the boundaries, the evolution equa-

tions should not contain terms involving the gradients of the variables. This

is a limitation of the theory as it excludes in particular the treatment of non-

local eﬀects. Some eﬀorts have been recently registered to circumvent this

diﬃculty but the problem is not deﬁnitively solved. Unlike extended irre-

versible thermodynamics, where great eﬀorts have been dedicated to a better

understanding of the notion of entropy and temperature outside equilibrium,

it seems that such questions are not of great concern in internal variables the-

ories. Here, the entropy that is used is the so-called accompanying entropy

and it is acknowledged that its rate of production is positive deﬁnite whatever

the number and nature of internal variables. The validity of such a hypothe-

sis if questionable and should be corroborated by microscopic theories as the

kinetic theory. The temperature is formally deﬁned as the derivative of the

internal energy with respect to entropy but questions about the deﬁnition

of a positive absolute temperature and its measurability in systems far from

equilibrium are even not invoked. It is expected that the validity of the re-

sults of the IVT become more and more accurate as the number of internal

234 8 Theories with Internal Variables

variables is increased and would become rigorously valid when the number

of variables is inﬁnite; however, from a practical point of view, this limit is

of course impossible to achieve.

8.4 Problems

8.1. Clausius–Duhem’s inequality. Show that the Clausius–Duhem’s inequal-

ity (8.10) is equivalent to the dissipation inequality (8.9).

8.2. Chemical reactions. Using the degree of advancement of a chemical reac-

tion ξ as an internal variable, formulate the problem of the chemical reaction

A + B = C + D in terms of the internal variable theory.

8.3. Particle suspensions. Why is the theory of molecular diﬀusion not ap-

plicable to the description of particle suspensions in ﬂuids?

8.4. Viscoelastic bodies. Derive the constitutive relation (8.26) of a Poynting–

Thomson body by using Liu’s Lagrange multiplier technique developed in

Chap. 9.

8.5. Colloidal suspensions. Establish the evolution equation (8.64) of the in-

ternal variable ξ by using (8.60), (8.61), and (8.63).

8.6. Colloidal suspensions. Derive (8.74) of the interaction force f between

the particles and the ﬂuid.

8.7. Colloidal suspensions. Eliminating the entropy ﬂux between (8.71) and

(8.73) show that the heat ﬂux in colloidal suspensions is given by

q = −λ∇T +

˜

hJ +η(u −v) ∇u.

In the particular problem of pure heat conduction, show that the above ex-

pression reduces to Fourier law, while for a mixture for which ρ

f

= ρ

p

(i.e.

u = v), it is equivalent to Dufour’s law.

8.8. Superﬂuids. Liquid He II is classically described by Landau’s two-ﬂuid

model (see for instance Khalatnikov 1965). Accordingly, He II is viewed as a

binary mixture consisting of a normal ﬂuid with a non-zero viscosity and a

superﬂuid with zero viscosity and zero entropy, the basic variables are ρ

n

, v

n

,

ρ

s

, v

s

, respectively, where ρ denotes the mass density and v the velocity ﬁeld.

Show that an equivalent description may be achieved by selecting the relative

velocity ξ = (ρ

n

/ρ)(v

n

−v

s

) as an internal variable, with the corresponding

Gibbs’ equation given by

T d(ρs) = d(ρu) −g dρ −αξ d(ρξ),

wherein ρ = ρ

n

+ ρ

s

, α = ρ

s

/ρ

n

while g = u − Ts + p(1/ρ) stands for the

speciﬁc Gibbs’ energy (see Lebon and Jou 1983; Mongiov`ı 1993, 2001).

8.4 Problems 235

8.9. Superﬂuids. Superﬂuid

4

He (see Lhuillier et al. 2003) is an ordered ﬂuid

of mass per unit volume ρ and momentum per unit volume ρv; the latter is

understood as the sum of two contributions: one from the condensate driving

the total mass and moving with velocity v

s

, the other from elementary ex-

citations of momentum p and zero mass: ρv = ρv

s

+ p. The other original

feature of the superﬂuid is that it manifests itself by a curl-free velocity:

∇v

s

= 0.

Following the reasoning of Sect. 8.2, establish that the evolution equation for

p, considered as an internal variable, is given by

∂p/∂t +∇ [(v +c)p] + [∇(v +c)] p = −ρs∇T −ρ∇ψ

D

−∇ τ

D

,

where c is the variable conjugate to p/ρ, i.e. [c = −T∂s/∂(p/ρ)], ψ

D

the

dissipative part of Gibbs’ function g = ψ +ψ

D

, τ

D

the dissipative part of the

mechanical stress tensor.

8.10. Continuous variable. The internal variable ξ can also take the form of

a continuous variable with a Gibbs’ equation written as

T ds = du −p dv −

_

µ(ξ)dρ(ξ)dξ.

If the rate of change of ρ(x) is governed by a continuity equation ∂ρ/∂t =

−∂J(ξ)/∂ξ, which deﬁnes J(ξ) as a ﬂux in the ξ-space, show that the corre-

sponding entropy production reads as

Tσ

s

= −

_

J(ξ)∂µ(ξ)/∂ξ dξ ≥ 0,

suggesting integral phenomenological relations. However, if the internal vari-

able does not change abruptly, it is suﬃcient to require that only the inte-

grand of the above expression is positive so that, J(ξ) = −L∂ρ(ξ)/∂ξ.

8.11. Application to Brownian motion. In this problem, the internal variable

ξ will be identiﬁed as the x-component u of the Brownian particle velocity

(ξ = u), and the density ρ(ξ) represents the velocity distribution which, at

equilibrium is the Maxwellian one,

f

eq

= constant exp(−mu

2

/2k

B

T).

Assume that the potential µ(u) is of the form

µ(u) = (k

B

T/m) ln ρ(u) +A(u),

where µ

eq

= µ

0

is independent of u. Combining the two previous relations,

determine the explicit expression of A(u). Show that the phenomenological

relation can be cast in the form J = −L[f(u) − (k

B

T/m)∂f/∂u], where L

236 8 Theories with Internal Variables

is the friction coeﬃcient of the Brownian particles. Combining this result

with the continuity relation, establish the Fokker–Planck equation for the

Brownian motion

∂f(u)

∂t

= L

_

∂f(u)

∂u

+

k

B

T

m

∂

2

f(u)

∂u

2

_

.

8.12. Magnetizable bodies. In theories of magnetic solids under strain, it is

customary to select magnetization M = B − H (with B the magnetic in-

duction and H the magnetic ﬁeld) as ﬁeld variable and to split the mag-

netic variables into a reversible and an irreversible contribution, for instance,

M = M

r

+ M

i

, H

r

+ H

i

. However, to describe the complex relaxation

process, some authors (Maugin 1999, p 242) have introduced an extra inter-

nal variable M

int

. With this choice, the entropy production takes the form

Tσ

s

= H

r

dM

r

dt

+H

i

dM

i

dt

+H

int

dM

int

dt

.

Show that H satisﬁes an evolution equation of the Cattaneo type

τ

dH

dt

+H=

τ

χ

m

dM

dt

,

where χ

m

denotes the magnetic susceptibility.

8.13. Vectorial internal variable and heat transport. Assume that the entropy

of a rigid heat conductor depends on the internal energy u and a vectorial in-

ternal variable j , i.e. s(u, j ). a) Obtain the constitutive equation for the time

derivative of j . b) From this equation, relate j to the heat ﬂux q and express

the evolution equation for q, assuming, for simplicity, that all phenomenolog-

ical coeﬃcients are constant; c) Compare this equation with the double-lag

equation presented in Problem 7.8. Which conditions are needed to reduce

it to the Maxwell-Cattaneo equation? Which form takes the entropy s(u, j )

when j is expressed in terms of q? Compare it with the extended entropy

(7.25).

Chapter 9

Rational Thermodynamics

A Mathematical Theory of Materials

with Memory

In Chaps. 7 and 8, it was assumed that the instantaneous local state of the

system out of equilibrium was characterized by the union of classical vari-

ables and a number of additional variables (ﬂuxes in EIT, internal variables

in IVT). Only their instantaneous value at the present time was taken into

account and their evolution was described by a set of ordinary diﬀerential

equations. An alternative attitude, followed in the early developments of ra-

tional thermodynamics (RT), is to select a smaller number of variables than

necessary for an exhaustive description. The price to be paid is that the state

of the material body will be characterized not only by the instantaneous value

of the variables, but also by their values taken in the past, namely by their

history.

In RT, non-equilibrium thermodynamic concepts are included in a contin-

uum mechanics framework. The roots of RT are found in the developments

of the rational mechanics. Emphasis is put on axiomatic aspects with theo-

rems, axioms and lemmas dominating the account. Coleman (1964) published

the foundational paper and the name “rational thermodynamics” was coined

a few years later by Truesdell (1968). RT deals essentially with deformable

solids with memory, but it is also applicable to a wider class of systems in-

cluding ﬂuids and chemical reactions. Its main objective is to put restrictions

on the form of the constitutive equations by application of formal statements

of thermodynamics. A typical feature of RT is that its founders consider it

as an autonomous branch from which it follows that a justiﬁcation of the

foundations and results must ultimately come from the theory itself. A vast

amount of literature has grown up about this theory which is appreciated

by the community of pure and applied mathematicians attracted by its ax-

iomatic vision of continuum mechanics.

In the present chapter we present a simpliﬁed “idealistic” but neverthe-

less critical version of RT laying aside, for clarity, the heavy mathematical

structure embedding most of the published works on the subject.

237

238 9 Rational Thermodynamics

9.1 General Structure

The basic tenet of rational thermodynamics is to borrow those notions and

deﬁnitions introduced in classical thermodynamics to describe equilibrium

situations and to admit a priori that they remain applicable even very far

from equilibrium. In that respect, temperature and entropy are considered as

primitive concepts which are a priori assigned to any state. Quoting Truesdell

(1984), it is suﬃcient to know that “temperature is a measure of how hot a

body is, while entropy, sometimes called the caloric, represents how much

heat has gone into a body from a body at a given temperature”.

Similarly, the second law of thermodynamics written in the form ∆S ≥

_

¯ dQ/T and usually termed the Clausius–Duhem’s inequality is always sup-

posed to hold. It is utilized as a constraint restricting the range of acceptable

constitutive relations. The consequence of the introduction of the history is

that Gibbs’ equation is no longer assumed to be valid at the outset as in the

classical theory of irreversible thermodynamics. Since the Gibbs equation is

abandoned, the distinction between state equations and phenomenological re-

lations disappears, everything will be collected under the encompassing word

of constitutive equations. Of course, the latter cannot take any arbitrary

forms as they have to satisfy a series of axioms, most of them being elevated

to the status of principles in the RT literature.

9.2 The Axioms of Rational Thermodynamics

To each material is associated a set of constitutive equations specifying partic-

ular properties of the system under study. In RT, these constitutive relations

take generally the form of functionals of the histories of the independent

variables and are kept distinct from the balance equations. In the present

chapter, the latter turn out to be

˙ ρ = −ρ∇ v (mass balance), (9.1a)

ρ ˙ v = ∇ σ +ρF (momentum balance), (9.1b)

ρ ˙ u = −∇ q +σ : ∇v +ρr (internal energy balance). (9.1c)

As in the previous chapters, a superimposed dot stands for the material

time derivative, ρ is the mass density; u, the speciﬁc internal energy; v, the

velocity; and q is the heat ﬂux vector; in rational thermodynamics, it is

preferred to work with the symmetric Cauchy stress tensor σ instead of the

symmetric pressure tensor P(= −σ). It is important to observe the presence

of the speciﬁc body force F in the momentum equation and the term r

in the energy balance, which represents the energy supply due to external

sources, for instance the energy lost or absorbed by radiation per unit time

9.2 The Axioms of Rational Thermodynamics 239

and unit mass. It must be realized that the body force and the source term

are essentially introduced for the self-consistency of the formalism. Contrary

to the classical approach, F and r are not quantities which are assigned a

priori, but instead the balance laws will be used to “deﬁne” them, quoting

the rationalists. In other terms, the balance equations of momentum and

energy are always ensured as we have two free parameters at our disposal.

The quantities F and r do not modify the behaviour of the body and do

not impose constraints on the set of variables, but rather, it is the behaviour

of the material, which determines them. This is a perplexing attitude, as F

and r, although supplied, will always modify the values of the constitutive

response of the system.

The principal aim of RT is to derive the constitutive equations character-

izing a given material. Of course, these relations cannot take arbitrary forms,

as they are submitted to a series of axioms, which place restrictions on them.

Let us brieﬂy present and discuss some of these most relevant axioms.

9.2.1 Axiom of Admissibility and Clausius–Duhem’s

Inequality

By “thermodynamically admissible” is understood a process whose constitu-

tive equations obey the Clausius–Duhem’s inequality and are consistent with

the balance equations. As will see later, the Clausius–Duhem’s inequality

plays a crucial role in RT. The starting relation is the celebrated Clausius–

Planck’s inequality, found in any textbook of equilibrium thermodynamics,

and stating that between two equilibrium sates A and B, one has

∆S ≥

_

B

A

¯ dQ/T. (9.2)

Since the total quantity of heat ¯ dQ results from the exchange with the exte-

rior through the boundaries and the presence of internal sources, the above

relation may be written as

d

dt

_

V

ρs dV ≥ −

_

Σ

1

T

q n dΣ +

_

V

ρ

r

T

dV , (9.3)

where s is the speciﬁc entropy, V is the total volume, and n is the outwards

unit normal to the bounding surface Σ. In local form, (9.3) writes as

ρ˙ s +∇

q

T

−

ρr

T

≥ 0. (9.4)

It is worth to note that the particular form (9.4) of the entropy inequality

is restricted to the class of materials for which the entropy supply is given

by ρr/T and the entropy ﬂux by q/T. For a more general expression of

240 9 Rational Thermodynamics

the entropy ﬂux, see extended irreversible thermodynamics (Chap. 7). After

elimination of r between the energy balance equation (9.1c) and inequality

(9.4) and introduction of Helmholtz’s free energy f(= u−Ts), one comes out

with

−ρ(

˙

f +s

˙

T) +σ : ∇v −q

∇T

T

≥ 0, (9.5)

which is referred to as the Clausius–Duhem’s or the fundamental inequality. It

is easily checked (see Problem 9.1) that the left-hand side of (9.5) represents

the rate of dissipated energy Tσ

s

per unit volume when the entropy ﬂux is

given by q/T.

9.2.2 Axiom of Memory

If it is admitted that the present is inﬂuenced not only by the present state

but also by the past history, the constitutive relations will depend on the

whole history of the independent variables. If ϕ(t) designates an arbitrary

function of time, say the temperature or the strain tensor, its history up to

the time t is deﬁned by ϕ

t

= ϕ(t −t

) with 0 ≤ t

< ∞.

The axiom of memory asserts that the behaviour of the system is com-

pletely determined by the history of the set of selected independent variables.

This means that the free energy, the entropy, the heat ﬂux and the stress

tensor, for instance, will be expressed as functionals of the history of the

independent variables. Considering the problem of heat conduction in a

rigid isotropic material, an example of constitutive equation with memory

is Fourier’s generalized law

q(t) =

_

t

−∞

λ(t −t

)∇T(t

)dt

, (9.6)

where λ(t − t

**) is the memory kernel. When this expression is substi-
**

tuted in the energy balance, one obtains an integro-diﬀerential equation

for the temperature ﬁeld, after use is made of ˙ u = c

˙

T with c the speciﬁc

heat capacity. If the memory kernel takes the form of an exponential like

(−λ/τ) exp[−(t − t

**)/τ], it is left as an exercise (see Problem 9.2) to show
**

that the time derivative of (9.6) is given by

τ ˙ q = −q −λ∇T, (9.7)

which is the same Cattaneo equation as in EIT. It is important to realize that

this result has been obtained by considering only the temperature as single

state variable. This is a characteristic of RT where the state space is generally

restricted to the classical variables, i.e. mass, velocity (or deformation), and

temperature, while the ﬂuxes are expressed in terms of integral constitutive

9.2 The Axioms of Rational Thermodynamics 241

equations containing the whole history of the independent variables. Instead

of assuming that q depends on the whole history of temperature ﬁeld, in

practical applications it is assumed that q is a function of ∇T and its higher-

order time derivatives. If the memory is very short in time, one may restrict

this sequence to a limited number of terms. But even in this case, RT oﬀers

an interesting formalism which departs radically from that of classical

irreversible thermodynamics.

In most situations, the notion of fading memory is also introduced. Accord-

ingly, the distant history has little inﬂuence on the present state; although

history is often described by an exponentially decreasing function of time,

it could take more general forms as a sum of exponentials or of Gaussian

memories. However, to avoid heavy mathematical developments, we shall

suppose from now on that the materials forget their past experience quasi-

instantaneously so that memory eﬀects can be neglected.

In some versions of RT, the description in terms of histories is substituted

by the state-process formalism (Noll 1974; Coleman and Owen 1974). Fol-

lowing these authors, to each thermodynamical system is associated a pair

formed by an instantaneous state and a process describing the temporal evo-

lution of the state space. The methods, tools and prospects of this theory

present similar features with these described in this chapter; an exhaustive

analysis of Coleman and Owen’s approach can also be found in Silhavy (1997).

9.2.3 Axiom of Equipresence

This axiom states that if a variable is present in one constitutive relation,

then there is no reason why it should not be present in all the other con-

stitutive equations, until it is proved otherwise. The condition for the pres-

ence or absence of an independent variable is essentially determined by the

Clausius–Duhem’s inequality. It should be realized that there is no physical

justiﬁcation to such an axiom, which is merely a mathematical convenience

in the determination of constitutive relations.

9.2.4 Axiom of Local Action

It is admitted that a material particle is only inﬂuenced by its immediate

neighbourhood and that it is insensitive to what happens at distant points.

Practically, it means that second and higher-order space derivatives are ex-

cluded from the constitutive relations. Higher spatial gradients have however

been included in some developments of the theory on non-local actions.

242 9 Rational Thermodynamics

9.2.5 Axiom of Material Frame-Indiﬀerence

Generally speaking, this axiom claims that the response of a system must

be independent of the motion of the observer. As the most general motion

of an observer, identiﬁed as a rigid coordinate system, is constituted by a

translation and a rotation, the axiom implies that the constitutive equations

must be invariant under the Euclidean transformation

x

∗

(t) = Q(t) x(t) +c(t). (9.8)

The quantity Q(t) is an arbitrary, real, proper orthogonal, time-valued tensor

satisfying

Q Q

T

= Q

T

Q = I, det Q = 1, (9.9)

c(t) is an arbitrary time-dependent vector; x(t), the position vector of a

material point at the present time and x

∗

(t) is the position occupied after

having undergone a rotation (ﬁrst term in the right-hand side of (9.8)) and

a translation (second term in the right-hand side). In the particular case

Q = I and c(t) = v

0

t with v

0

a constant velocity, (9.8) reduces to a Galilean

transformation.

When the Euclidean group (9.8) acts on a tensor of rank n(n = 0, 1, 2),

the latter is said to be objective if it transforms according to

a

∗

= a (objective scalar), (9.10a)

a

∗

= Q a (objective vector), (9.10b)

A

∗

= Q A Q

T

(objective tensor). (9.10c)

We directly observe that the velocity v

∗

= dx

∗

/dt is not an objective vector

as it transforms as

v

∗

= Q v +

˙

Q x + ˙ c,

which is not of the form (9.10b) because of the presence of the second and

third terms in the right-hand side of the above relation; the same is true for

the acceleration, which is not objective. It is easily checked (see Problems 9.3

and 9.4) that the velocity gradient, the angular velocity tensor and the ma-

terial time derivative of objective vectors or tensors are non-objective; in

contrast, quantities like the temperature gradient, the mass density gradient

or the symmetric velocity gradient tensor are objective.

We are now in position to propose a more precise formulation of the princi-

ple of material frame-indiﬀerence which rests on the following requirements.

First, the primitive variables as temperature, energy, entropy, free energy

and energy supply are by essence objective scalars, the body force and the

heat ﬂux are objective vectors while the stress tensor is an objective ten-

sor. Second, the constitutive relations are objective, i.e. form invariant with

respect to the Euclidean transformation (9.8). Third, the constitutive equa-

tions, which reﬂect the material properties of a body, cannot depend on the

9.3 Application to Thermoelastic Materials 243

angular velocity of the reference frame. To give an example, the Newton’s

equation of rational mechanics, when formulated in a non-inertial rotating

frame, is Euclidean invariant but it depends explicitly on the angular ve-

locity of the frame and therefore, it does not fulﬁl the axiom of material

frame-indiﬀerence.

9.3 Application to Thermoelastic Materials

Consider a deformable, anisotropic elastic solid. Under the action of exter-

nal forces and heating, the material changes its conﬁguration from a non-

deformed one with mass density ρ

0

to a deformed state with mass density

ρ. The position of the material points is denoted by X, in the non-deformed

conﬁguration and by x = χ(X, t), in the deformed one with u = x −X the

displacement vector.

Loyal to our principle of simplicity, we shall restrict the analysis to linear

thermoelasticity, i.e. small deformations and small temperature increments

T −T

0

with respect to a reference temperature T

0

. A more general description

implying large deformations is treated in Box 9.1. Within the limit of small

deformations, the density ρ remains constant and the balance equations for

the displacement vector u and the temperature ﬁeld read as (Eringen 1967;

Truesdell and Toupin 1960)

ρ¨ u = ∇ σ +ρF, (9.11)

ρ ˙ u = −∇ q +σ : ˙ ε +ρr, (9.12)

where ε =

1

2

[∇u + (∇u)

T

] stands for the symmetric strain tensor. These

relations contain unknown quantities as u (or equivalently f), σ and q

which must be speciﬁed by constitutive equations, compatible with Clausius–

Duhem’s inequality. In (9.12), the scalar u (internal energy) should not be

confused with the vector u (displacement vector) appearing in (9.11).

A thermoelastic material is deﬁned by the following constitutive equations

f = f(T, ∇T, ε), (9.13)

s = s(T, ∇T, ε), (9.14)

σ = σ(T, ∇T, ε), (9.15)

q = q(T, ∇T, ε), (9.16)

where use has been made of the axiom of equipresence, s has been included

among the constitutive relations as it ﬁgures explicitly in Clausius–Duhem’s

relation. By no means are the above constitutive equations the most general

that one could propose but they appear as particularly useful to describe a

large class of deformable elastic solids. As a consequence of (9.13) and using

the chain diﬀerentiation rule, one can write the time derivative of f as

244 9 Rational Thermodynamics

˙

f =

∂f

∂T

˙

T +

∂f

∂(∇T)

.

∇T +

∂f

∂ε

: ˙ ε. (9.17)

It is left as an exercise (see Problem 9.5) to prove that the Clausius–Duhem’s

inequality (9.5) will take the form

−ρ(

˙

f +s

˙

T) +σ : ˙ ε −

q ∇T

T

≥ 0, (9.18)

and, after substitution of (9.17),

−ρ

_

s +

∂f

∂T

_

˙

T −

_

ρ

∂f

∂ε

−σ

_

: ˙ ε −ρ

∂f

∂∇T

.

∇T −q

∇T

T

≥ 0, (9.19)

which is obviously linear in

˙

T, ˙ ε, and

·

∇T. Moreover, since there exist body

forces and energy supplies that ensure that the balance equations of momen-

tum and energy are identically satisﬁed, these laws do not impose constraints

on

˙

T, ˙ ε, and

·

∇T, which can therefore take arbitrary prescribed values. In or-

der that the entropy inequality (9.19) holds identically, it is then necessary

and suﬃcient that the coeﬃcient of each time derivative vanishes. As a con-

sequence, it follows that

s = −∂f/∂T, (9.20a)

σ = ρ∂f/∂ε, (9.20b)

∂f/∂(∇T) = 0. (9.20c)

It is concluded from (9.20c) that the free energy f does not depend on the

temperature gradient and on account of (9.20a) and (9.20b), the same obser-

vation holds for the entropy s and the stress tensor σ so that (9.13)–(9.15)

will take the form

f = f(T, ε), s = s(T, ε), σ = σ(T, ε). (9.21)

From (9.20a), (9.20b), and (9.20c), we can write the diﬀerential expression

df = −sdT +ρ

−1

σ : dε (9.22)

or, equivalently,

T ds = du −ρ

−1

σ : dε, (9.23)

which is the Gibbs equation for thermoelastic bodies. Note that this relation

has not been assumed as a starting point but has been derived within the

formalism.

Expanding f up to the second order in σ and T − T

0

, one obtains, in

Cartesian coordinates, and using Einstein’s summation convention,

f =

1

2ρ

C

ijkl

ε

ij

ε

kl

−

1

2

c

ε

T

0

(T −T

0

)

2

−

1

ρ

β

ij

ε

ij

(T −T

0

), (9.24)

9.3 Application to Thermoelastic Materials 245

where C

ijkl

is the fourth-order tensor of elastic moduli; c

ε

, the heat capacity

and β

ij

is the second-order tensor of thermal moduli. In virtue of (9.20a)

and (9.20b), the corresponding linear constitutive equations of s and σ

ij

are

given by

s =

c

ε

T

0

(T −T

0

) +

1

ρ

β

ij

ε

ij

, (9.25)

σ

ij

= C

ijkl

ε

kl

−β

ij

(T −T

0

). (9.26)

The result (9.26) is the well-known Neumann–Duhamel’s relation of ther-

moelasticity, which simpliﬁes to Hooke’s law when the temperature is uniform

(T = T

0

).

Going back to Clausius–Duhem’s expression (9.19), the latter reduces to

the remaining inequality

q(T, ∇T, ε) ∇T ≤ 0, (9.27)

which reﬂects the property that heat ﬂows spontaneously from high to low

temperatures. Deﬁning the heat conductivity tensor by

K(T, ε) = −

_

∂q(T, ∇T, ε)

∂(∇T)

_

∇T=0

(9.28)

and expanding q around ∇T = 0 with T and ε ﬁxed, one obtains in the

neighbourhood of ∇T = 0, i.e. by omitting non-linear terms,

q(T, ∇T, ε) = q(T, 0, ε) −K(T, ε) ∇T. (9.29)

Substitution of (9.29) in (9.27) yields,

q(T, 0, ε) ∇T −∇T K(T, ε) ∇T ≤ 0. (9.30)

Since this relation must be satisﬁed for all ∇T, it is required that

q(T, 0, ε) = 0, ∇T K(T, ε) ∇T ≥ 0. (9.31)

From this, it is immediately concluded that in an admissible thermoelastic

process, the heat ﬂux is zero when the temperature gradient vanishes and the

heat conductivity tensor, which is independent of ∇T, is positive deﬁnite. In

fact, only the symmetric part K

sym

= K + K

T

of K enters the statement

(9.31b) to be positive deﬁnite, this is so because for the skew part K

skew

=

K−K

T

, one has ∇T K

skew

∇T = 0. It follows from the above considerations

that expansion (9.29) becomes

q(T, ∇T, ε) = −K(T, ε) ∇T. (9.32)

This is a generalization of Fourier’s equation in which the thermal conduc-

tivity is a function not only of the temperature but also of the strain tensor.

246 9 Rational Thermodynamics

We observe also from (9.32) that in a thermoelastic solid, it is not possible

to produce a heat ﬂux by a deformation only, at uniform temperature. This

means that the presence of a piezoelectric eﬀect is excluded in thermoelastic

bodies.

The generalization of the above results to large deformations is straight-

forward and is presented in the Box 9.1.

Box 9.1 Finite Deformations in Thermoelastic Materials

It is usual in solid mechanics to introduce the following quantities (e.g.

Eringen 1962):

F = ∇χ (deformation tensor, F

ij

= ∂x

i

/∂X

j

in Cartesian coordi-

nates, with x

i

and X

j

the position in the deformed and non-

deformed reference conﬁguration, respectively),

E =

1

2

(F

T

F −I) (Green symmetric strain tensor),

L =

˙

F F

−1

= ∇v (strain rate tensor, (∇v)

ij

= ∂v

j

/∂x

i

in

Cartesian coordinates),

V = symL (symmetric part of L),

T = (ρ

0

/ρ)σ F

−T

(ﬁrst Piola–Kirchhoﬀ stress tensor),

S = F

−1

T (symmetric second Piola–Kirchhoﬀ stress tensor),

Q = (ρ

0

/ρ)F

−1

q (heat ﬂux measured in the reference non-

deformed conﬁguration).

The balance laws of mass, momentum and energy take the form

ρ

0

/ρ = det F,

ρ

0

˙ v = ∇ T+ρ

0

F,

ρ

0

˙ u = −∇ q +S :

˙

E+ρr.

The results derived for small deformations are still valid at the condition

to replace everywhere ε by E, σ by S, q by Q, and ∇T by G = F

T

∇T,

the temperature gradient with respect to the reference conﬁguration. For

example, the Clausius–Duhem’s inequality reads as

−ρ

0

_

∂f

∂T

+s

_

˙

T −

_

ρ

0

∂f

∂E

−S

_

˙

E−ρ

0

∂f

∂G

˙

G −Q

G

T

≥ 0,

leading to the restrictions,

∂f/∂T = −s, ∂f/∂E = S/ρ

0

, ∂f/∂G = 0,

with a generalized Fourier law given by

Q(T, G, E) = −K(T, E) G.

9.4 Viscous Heat Conducting Fluids 247

Although most of the applications of RT have been devoted to solid mechan-

ics, there is no problem to adapt the above considerations to hydrodynamics,

as shown in Sect. 9.4.

9.4 Viscous Heat Conducting Fluids

To avoid lengthy mathematical developments with constitutive equations in

functional form, we shall consider a particular class of ﬂuids characterized by

isotropy, absence of memory and described by the following set of constitutive

equations:

f = f(v, T, V, ∇T), (9.33)

s = s(v, T, V, ∇T), (9.34)

σ = σ(v, T, V, ∇T), (9.35)

q = q(v, T, V, ∇T), (9.36)

wherein v = ρ

−1

is the speciﬁc volume, V =

1

2

[∇v + (∇v)

T

] the symmetric

velocity gradient tensor. Note that neither the velocity ﬁeld v nor the ve-

locity gradient ∇v have been included in the set of independent variables,

because they are not objective quantities and therefore do not satisfy the

axiom of frame-indiﬀerence. Instead, the tensor V has been selected as it

meets the property to be objective. Observe in passing that the above re-

sponse functions are in full agreement with the axiom of equipresence. More

restrictions on the constitutive relations are placed by Clausius–Duhem’s in-

equality (9.5). Introducing (9.33)–(9.36) in inequality (9.5) and applying the

chain diﬀerentiation rule to calculate

˙

f, one obtains

−ρ

_

s +

∂f

∂T

_

˙

T−ρ

∂f

∂V

:

˙

V−ρ

∂f

∂(∇T)

·

(∇T)−

1

T

q∇T−

_

∂f

∂v

I −σ

_

: V ≥ 0,

(9.37)

wherein use has been made of the continuity equation ρ˙ v = ∇ v = V : I.

On looking at inequality (9.37), we see that it is a linear expression in the

time derivatives

˙

T,

˙

V, and

·

(∇T); if it is assumed that there are body forces

and energy supply that ensure that the momentum and energy balance equa-

tions are identically satisﬁed, one can assign to these derivatives arbitrary

and independent values. Clearly inequality (9.37) will not hold unless the

coeﬃcients of these derivatives are zero, which leads to the following results:

s = −

∂f

∂T

, (9.38)

∂f

∂V

= 0, (9.39)

∂f

∂(∇T)

= 0. (9.40)

248 9 Rational Thermodynamics

Relation (9.38) is classical and from the next ones, it is deduced that the free

energy f (and as a corollary the entropy s) is a function of v and T alone so

that

f = f(v, T), s = s(v, T). (9.41)

Deﬁning the equilibrium pressure p by p = −∂f/∂v, which is justiﬁed as p

can only depend on the “equilibrium” variables v and T, and combining with

(9.38), one ﬁnds the following Gibbs equation:

df = −s dT −p dv. (9.42)

It is important to notice that, in contrast with classical irreversible thermo-

dynamics where the Gibbs relation is postulated from the outset, in RT it is

a derived result.

Furthermore, in virtue of the results (9.38)–(9.40), the Clausius–Duhem’s

inequality reduces to

_

−

∂f

∂v

I +σ

_

: V−

1

T

q ∇T ≥ 0. (9.43)

Introducing the viscous stress tensor σ

(v)

deﬁned by

σ

(v)

(v, T, V, ∇T) = σ(v, T, V, ∇T) +p(v, T)I, (9.44)

(9.43) is written as

σ

(v)

: V−q

∇T

T

≥ 0, (9.45)

and represents the rate of energy dissipated per unit volume of the ﬂuid.

Explicit expressions for the constitutive equations for q and σ

(v)

are di-

rectly obtained by using the representation theorems for isotropic tensors

(Truesdell and Toupin 1960). In the linear approximation, when second and

higher-order terms in ∇T and V are omitted, it is left as an exercise (Prob-

lem 9.7) to prove that

q = −λ(v, T)∇T, σ

(v)

= γ(v, T)(∇ v)I + 2η(v, T)V. (9.46)

One recognizes the Fourier law with λ the heat conductivity coeﬃcient and

the Newton’s law of hydrodynamics, with η the dynamical shear viscosity

and the coeﬃcient γ related to the bulk viscosity; of course, the scalars λ, γ,

and η are all three functions of v and T. The ﬂuids described by the linear

constitutive relations (9.46) are usually referred to as Fourier–Stokes–Newton

ﬂuids. After substitution of (9.46) in (9.45), it is observed that satisfaction

of Clausius–Duhem’s relation yields the following inequalities, which are well

known in ﬂuid mechanics:

λ > 0, γ +

2

3

η > 0, η > 0. (9.47)

Summarizing, we can assert that, for the class of ﬂuids described by constitu-

tive equations of the formφ = φ(v, T, ∇T, V) with φ ≡ f, s, q, σ, respectively:

9.5 Comments and Critical Analysis 249

1. The thermodynamic potentials f and s do not depend on ∇T and V, so

that f = f(v, T) and s = s(v, T).

2. The classical Gibbs equation (9.42) or equivalently Tds = du + p dv is

demonstrated to remain valid.

3. In the linear approximation, the constitutive equations for q and σ

(v)

are

the traditional equations of Fourier and Stokes–Newton, respectively.

4. As a side result of Clausius–Duhem’s inequality, the heat conductivity

and the viscosity coeﬃcients are shown to be positive.

Despite the fact that rational thermodynamics radiates some taste of el-

egance and generality, it has been the subject of acrid criticisms, which are

discussed in Sect. 9.5.

9.5 Comments and Critical Analysis

The axiomatic approach of RT has been the subject of severe critical obser-

vations (e.g. Lavenda 1979; Woods 1981; Rivlin 1984) for the lack of physical

background and touch with experiments. In what follows, we shall shortly

discuss the most frequently criticisms addressed against the formalism.

9.5.1 The Clausius–Duhem’s Inequality

The basic idea is to use the Clausius–Duhem’s inequality to place restrictions

on the form of the constitutive equations. However, the original formulation

of Clausius–Duhem’s inequality is restricted to processes taking place be-

tween two equilibrium states. In RT, it is applied, without any justiﬁcation,

to arbitrary processes taking place between non-equilibrium states.

Furthermore, the local Clausius–Duhem’s inequality amounts to admit

that the positiveness of the entropy production σ

s

is a necessary and suﬃ-

cient condition allowing to restrict the range of acceptable constitutive rela-

tions. However, it is not proved that σ

s

≥ 0 remains valid when truncated

expressions of σ

s

are used, as is generally the case in RT. By performing

a series expansion as σ

s

= σ

1

+ σ

2

+ σ

3

+ , where σ

j

(j = 1, 2, 3, . . .) is

the entropy production at the j order, it is absolutely not ensured that the

entropy production remains positive deﬁnite at any order of approximation.

9.5.2 Axiom of Phlogiston

The term phlogiston is borrowed from Woods (1981) and it designates a neg-

ative mass medium permeating all bodies and expelled by heat. The axiom

250 9 Rational Thermodynamics

concerns the property that in any process, time derivatives such as

˙

T,

·

∇T,

˙

V can be given arbitrary and independent values of T, ∇T, and V. In nor-

mal circumstances, in the balance equations of momentum and energy, the

body force F and the source term r are known and take well-speciﬁed values:

the deformation (or velocity) and temperature ﬁelds are then determined by

solving the equations after that initial and boundary conditions are given. In

RT, the procedure is reversed: for any value of the velocity v (or deformation

u) and the temperature T ﬁelds, it is admitted that one can select appropri-

ately F and r so that the equations of momentum and energy are identically

satisﬁed and do not impose any restriction on the set of variables. In a real

experience, F and r are speciﬁed by physical rules, which are beyond our

control and therefore they cannot be speciﬁed throughout the medium and

in that respect, it can be said that the phlogiston axiom destroys the empiri-

cal content of the balance equations of momentum and energy. The presence

of quantities F and r is required to guarantee that the state variables and

their time derivatives can be varied independently. This limits seriously the

domain of applicability of the theory as it fails to describe processes for which

the variables are not independent of their time variations.

The problems linked with the use of the phlogiston axiom can be however

circumvented thanks to an elegant technique proposed by Liu (1972) and

outlined in Appendix 1.

9.5.3 The Meaning of Temperature and Entropy

It is important to realize that in a majority of works on RT, entropy and

temperature are considered as primitive undeﬁned objects and their physical

meaning is not a subject of deep concern. Regarding the entropy, it is simply

given by a constitutive relation expressing its functional dependence with re-

spect to the selected variables while the temperature T remains an undeﬁned

variable which is speciﬁed by the laws it satisﬁes. Quoting Truesdell (1968),

“As for physical meaning, I claim no physical applicability for anything I

ever say. . . Whether a theory applies to a given piece of material at a given

time is something very important, but something that the theorist cannot be

expected to tell, in thermodynamics or any other theory.” It must however

be added that in the later developments (e.g. Day 1972; Coleman and Owen

1974; Serrin 1979; Coleman et al. 1981; Kratochvil and Silhavy 1982), more

attention has been paid to prove the existence of an entropy as a state func-

tion as well as an absolute non-equilibrium temperature. The existence of

entropy and absolute temperature is no longer postulated but deduced from

statements involving cyclic processes; nevertheless, the fundamental problem

of an unambiguous deﬁnition of temperature and entropy outside equilibrium

remains called into question. For example, some people have questioned the

measurability of the variable T introduced in the theory: a priori there should

9.5 Comments and Critical Analysis 251

be no reason to identify this T with the temperature given by a gas thermome-

ter or a thermocouple. Recent investigations (Muschik 1977; Casas-V´ azquez

and Jou 2003; Crisanti and Ritort 2003) have brought out that a precise deﬁ-

nition of temperature and entropy is of prime importance in non-equilibrium

thermodynamics. In addition, it was proved that the entropy used in rational

thermodynamics is not unique. Meixner (1973a, 1973b) has shown that there

exists an inﬁnity of functionals, all deserving the name of entropy, that pos-

sess the property to satisfy the Clausius–Duhem’s inequality and that there

is no criterion that permits to favour one deﬁnition over the others. Meixner’s

arguments were reinforced by a more particular analysis by Day (1977) who

demonstrated the non-uniqueness of entropy in the case of heat conduction

in a rigid isotropic body with memory. Another example of a system having

many diﬀerent types of entropy was given by Coleman and Owen (1975) and

concerns elastic–plastic materials. The failure of the entropy to be not unique

is not eliminated in the state-process version of the theory.

9.5.4 Axiom of Frame-Indiﬀerence

This axiom, that requires the invariance of constitutive relations under time-

dependent accelerations and rotations of the actual reference frame, has

been the subject of intense debate and controversy. As ﬁrst observed by

M¨ uller (1972) and Edelen and McLennan (1973), the axiom excludes phys-

ical processes which are Galilean invariant. Frame-indiﬀerence is not sat-

isﬁed in a whole series of disciplines; in particular in rational mechanics.

Newton’s law of motion formulated in a non-inertial system is “objective”

but not frame dependent as the non-inertial forces depend explicitly on the

angular velocity of the reference frame; in the kinetic theory of gases, the

Burnett equations generalizing the classical Fourier–Stokes–Newton’s laws

have also been shown to be frame dependent; in the theory of turbulence,

it is experimentally observed that the turbulent viscosity takes diﬀerent val-

ues according reference frame is inertial or non-inertial (Lumley 1983). Fur-

ther violations of material indiﬀerence are found in rheology (Bird and de

Gennes 1983) and molecular hydrodynamics (Hoover et al. 1981). Another

illustration of non-respect of frame-indiﬀerence is provided by the classical

theory of irreversible processes: referring to Chap. 2, we know that the phe-

nomenological coeﬃcients L

αβ

are depending on the angular velocity when

measured in a rotating frame. It can be said that actually, there are serious

evidences against the universality of the axiom of frame-indiﬀerence which

has to be regarded as “a convenience rather than a principle”, as concluded by

Edelen and McLennan (1973). Notwithstanding, axiom of frame-indiﬀerence

is widely adopted throughout rational thermodynamics and has disturbed

people working in continuum mechanics. Fortunately, most results of con-

tinuum mechanics are established in inertial frames so that the eﬀects of

non-inertial forces can be ignored.

252 9 Rational Thermodynamics

9.5.5 The Entropy Flux Axiom

It is also largely admitted in rational thermodynamics that the entropy ﬂux

is given by J

s

= q/T (plus eventually a term −µJ/T in presence of matter

diﬀusion) as one may expect from the Clausius–Planck’s formula. Although

this result is correct in the linear approximation, this is certainly not true

in higher-order theories as conﬁrmed by the kinetic theory of gases (Grad

1958) or extended irreversible thermodynamics. Referring to Chap. 7, for a

wide class of processes, the entropy ﬂux for pure substances is not the heat

ﬂux divided by the temperature but of the more general form

J

s

=

q

T

−

1

3

β

(tr σ

v

)q −β

0

σ

v

q, (9.48)

wherein β

and β

**are phenomenological coeﬃcients and
**

0

σ

v

is the deviatoric

part of the viscous stress tensor σ

v

. This result is one of the most impor-

tant diﬀerences between rational thermodynamics and extended irreversible

thermodynamics.

9.5.6 The Axiom of Equipresence

This axiom states that all constitutive relations depend precisely on the same

set of variables, unless it is proved otherwise. As a matter of fact, the con-

clusion always shows that the response functions do not depend generally on

the whole set of variables. Of course, there is no physical argument for such

an axiom, which is essentially a mathematical convenience while formulat-

ing constitutive equations. It is certainly overweening to elevate equipresence

to the status of axiom or principle because it represents merely a technical

commodity.

To conclude, RT is an axiomatic theory characterized by generality and

mathematical elegance. Unfortunately, mathematical rigour has been ob-

tained at the detriment of physical insight and this explains some lack of

interest from some corporations of physicists and engineers. Nevertheless, it

is our opinion that it is a theory that deserves a close attention. As mentioned

earlier, some of the criticisms should be moderated and some of them are even

avoidable, for instance the phlogiston can be circumvented by appealing to

Liu’s technique while other criticisms, like these addressed again the axiom

of equipresence, are of minor consequence. It was the merit of RT to be the

ﬁrst non-equilibrium formalism to get rid of the local equilibrium hypoth-

esis and to go beyond the linear regime to which is restricted the classical

theory of non-equilibrium processes. RT has been applied to a huge number

of problems mainly in the ﬁelds of non-linear elasticity, coupled mechanical,

thermal and electro-magnetic phenomena, rheology, wave propagation, and

9.5 Comments and Critical Analysis 253

shock waves. Most of them are published in the “Archive of Rational Mechan-

ics and Analysis” which is the privileged tribune of “rationalists”. Moreover,

the tools of RT have been widely applied in other formalisms as rational

extended thermodynamics (M¨ uller and Ruggeri 1998), theories with internal

variables (Maugin 1999) or GENERIC (

¨

Ottinger 2005).

Appendix 1: Liu’s Lagrange Multipliers

An elegant alternative to the admissibility axiom of rational thermodynamics,

i.e. the necessary and suﬃcient conditions to satisfy the Clausius–Duhem’s

inequality, was proposed by Liu (1972). He was able to show that the entropy

inequality (9.5) becomes valid for completely arbitrary variations of the vari-

ables at the condition to consider the balance equations of mass, momentum

and energy as mathematical constraints. To be more explicit, each balance

equation is multiplied by an appropriate factor, named Lagrange multiplier

by analogy with the extremization problem in mathematics, and the resulting

vanishing quantity is added to the left-hand side of the entropy inequality

(9.5). Let us illustrate the procedure by means of the simple example of heat

conduction in a rigid isotropic body without source term. The behaviour of

the temperature ﬁeld is governed by one single ﬁeld equation, namely the

energy balance

ρ ˙ u = −∇ q, (9.49)

it being understood that there are restrictions placed by the entropy inequality

ρ˙ s +∇

_

q

T

_

≥ 0. (9.50)

Closure relations are provided by the following set of constitutive equations:

u = u(T, ∇T), s = s(T, ∇T), q = q(T, ∇T). (9.51)

The requirement that the temperature ﬁeld satisfying the entropy inequality

(9.50) must also be a solution of the energy balance equation (9.49) is inter-

preted by Liu as a constraint. It was proved by Liu that this constraint can

be eliminated by the introduction of Lagrange multipliers and by writing the

entropy inequality under the new form

ρ˙ s +∇

_

q

T

_

−Λ(ρ ˙ u +∇ q) ≥ 0; (9.52)

the quantity Λ designates the Lagrange multiplier, which depends generally

on the whole set of variables and must be determined from the formalism.

After that the constitutive relations (9.51) are inserted in inequality (9.52)

and all diﬀerentiations are performed, one obtains a relation that is linear in

the arbitrary derivatives

˙

T,

·

∇T, ∇(∇T):

254 9 Rational Thermodynamics

ρ

_

∂s

∂T

−Λ

∂u

∂T

_

˙

T +ρ

_

∂s

∂(∇T)

−Λ

∂u

∂(∇T)

_

(

·

∇T)

+

_

1

T

−Λ

_

∂q

∂(∇T)

: ∇(∇T) +

__

1

T

−Λ

_

∂q

∂T

−

1

T

2

q

_

∇T ≥ 0. (9.53)

Since this inequality could be violated unless the coeﬃcients of these deriva-

tives vanish, it is found that:

∂s

∂T

−Λ

∂u

∂T

= 0,

∂s

∂(∇T)

−Λ

∂u

∂(∇T)

= 0, Λ =

1

T

. (9.54)

An important result is the third relation from which it is concluded that

the Lagrange multiplier can be identiﬁed as the inverse of the temperature.

Moreover if it is recalled that the free energy is deﬁned as f = u − Ts, it

follows from the ﬁrst and the second relation (9.54) that

∂f

∂T

= −s,

∂f

∂(∇T)

= 0. (9.55)

This implies that f is independent of ∇T and that f satisﬁes the classical

Gibbs relation df = −sdT. Another consequence of (9.55) is that the entropy

s is also independent of the temperature gradient; this property is also shared

by the internal energy u, in virtue of the deﬁnition of f.

There still remains from (9.53) the residual inequality

q ∇T ≤ 0. (9.56)

Clearly, the simplest way to guarantee that (9.56) is negative deﬁnite is to

assume that

q = −λ(T)∇T with λ ≥ 0, (9.57)

and we are back with Fourier’s law. From now on, the procedure is classical.

Replacing (9.57) in the energy balance equation (9.49) and writing for u(T) a

constitutive equation such that ˙ u = c

˙

T, with c the speciﬁc heat capacity, one

is led to the diﬀusion equation c

˙

T = ∇ (λ∇T) which, after that initial and

boundary conditions are speciﬁed, allows us to determine the temperature

distribution in the body.

Appendix 2: Rational Extended Thermodynamics

The formulation of EIT, in Chap. 7, was inspired by the concepts and methods

of CIT. But EIT may equivalently be described by making use of the tools

and structure of rational thermodynamics: it is then referred to as rational

extended thermodynamics.

As an illustration, consider the problem of hyperbolic heat conduction in

a rigid isotropic body, in absence of an energy source term. The relevant

9.5 Comments and Critical Analysis 255

variables are the internal energy u and the heat ﬂux q. The time evolution of

u is governed by the balance law of energy (9.49) while the evolution equation

of q will be cast in the general form

ρ ˙ q = −∇ Q+σ

q

, (9.58)

where Q (a second-order tensor) denotes the ﬂux of the heat ﬂux and σ

q

(a

vector) is a source term. These quantities must be formulated by means of

constitutive equations which, for simplicity, will be given by

Q = Q(u, q) = a(u, q

2

)I, σ

q

= σ

q

(u, q) = b(u, q

2

)q, (9.59)

wherein the scalars a(u, q

2

) and b(u, q

2

) are unknown functions of u and q

2

to be determined.

Following Liu’s technique, the entropy inequality will be formulated in

such a way that the constraints imposed by the energy balance and (9.58)

are explicitly introduced via the Lagrange multipliers Λ

0

(u, q) (a scalar) and

Λ

1

(u, q) (a vector), so that

ρ˙ s +∇ J

s

−Λ

0

(ρ ˙ u +∇ q) −Λ

1

(ρ ˙ q +∇ Q−σ

q

) ≥ 0, (9.60)

where s(u, q

2

) and J

s

(u, q) are arbitrary functions of u and q. By diﬀerenti-

ating s and J

s

with respect to u and q, and rearranging the various terms,

one may rewrite (9.60) as:

ρ

_

∂s

∂u

−Λ

0

_

˙ u +ρ

_

2

∂s

∂q

2

q −Λ

1

_

˙ q +

∂J

s

∂u

∇u +

∂J

s

∂q

: ∇q

−Λ

0

∇ q −Λ

1

∇u

∂a

∂u

−2Λ

1

q : ∇q

∂a

∂q

2

+bΛ

1

q ≥ 0. (9.61)

Since inequality (9.61) is linear in the arbitrary derivatives ˙ u, ˙ q, ∇u, ∇q,

positiveness of (9.61) requires that their respective factors vanish, from which

it results that

∂s

∂u

= Λ

0

(≡ θ

−1

), 2

∂s

∂q

2

q = Λ

1

(≡ γ(u, q

2

)q) (9.62)

and

∂J

s

∂u

=

∂a

∂u

Λ

1

,

∂J

s

∂q

= Λ

0

I + 2Λ

1

q

∂a

∂q

2

, (9.63)

wherein we have identiﬁed Λ

0

with θ

−1

, the inverse of a non-equilibrium

temperature (see Box 7.4) and, without loss of generality, Λ

1

with γ(u, q

2

)q

where γ is an arbitrary function of u and q

2

.

Taking into account of the results (9.62) and (9.63), the entropy inequality

(9.61) reduces to the residual inequality

bΛ

1

q ≥ 0. (9.64)

256 9 Rational Thermodynamics

The fact that the entropy ﬂux is an isotropic function implies that

J

s

= ϕ(u, q

2

)q, (9.65)

which, substituted in (9.63), yields

∂ϕ

∂u

= γ

∂a

∂u

(a) and (ϕ −θ

−1

)I + 2

_

∂ϕ

∂q

2

−γ

∂a

∂q

2

_

qq = 0 (b). (9.66)

Since the dyadic product qq is generally not zero, it follows that

∂ϕ

∂q

2

= γ

∂a

∂q

2

(a) and ϕ = θ

−1

(b). (9.67)

The second result (9.67) is important as it indicates that ϕ is equal to the

inverse of the temperature and as a consequence that the entropy ﬂux (9.65)

is given by the usual expression

J

s

= q/θ. (9.68)

Moreover, in virtue of (9.66a) and (9.66b), one has

dθ

−1

≡ dϕ =

∂ϕ

∂u

du +

∂ϕ

∂q

2

dq

2

= γda, (9.69)

a result that will be exploited to obtain the ﬁnal expression of the evolution

equation of the heat ﬂux. Indeed, after making use of the results σ

q

= bq

and ∇ Q = ∇a = γ

−1

∇θ

−1

in (9.58), it is found that

ρ ˙ q = −

1

γ

∇θ

−1

+bq. (9.70)

Dividing both members of (9.70) by b and setting ρ/b = −τ, (γb)

−1

= L, one

recovers a Cattaneo-type relation

τ ˙ q = L∇θ

−1

−q, (9.71)

wherein τ can be identiﬁed as a relaxation time and L as a generalized heat

conductivity. In terms of these coeﬃcients, the Lagrange multiplier Λ

1

is

given by

Λ

1

= γq = −(τ/ρL)q. (9.72)

Let us ﬁnally derive the Gibbs equation, it follows directly from (9.62) that

ds =

∂s

∂u

du +

∂s

∂q

2

dq

2

=

∂s

∂u

du +Λ

1

dq = θ

−1

du −

_

τ

ρL

_

q dq, (9.73)

where the identiﬁcation (9.72) has been introduced. It is worth to stress that

this relation is similar to (7.24). Equality of the second-order derivatives

9.6 Problems 257

of s in (9.73) leads to the Maxwell relation ∂θ

−1

/∂q

2

=

1

2

(∂γ/∂u) and, after

integration,

θ

−1

(u, q

2

) = T

−1

(u) +

1

2

_

∂γ(u, q

2

)

∂u

dq

2

, (9.74)

where T(u) is the local equilibrium temperature. Clearly, the non-equilibrium

temperature reduces to its equilibrium value when the factor γ (or equiva-

lently the Lagrange multiplier Λ

1

) is independent of u.

We close this analysis with some considerations about the sign of the

various coeﬃcients. Stability of (local) equilibrium requires that the second

variation of entropy with respect to the state variables u and q is negative

deﬁnite, so that in particular,

∂

2

s

∂q

2

= γ ≤ 0, (9.75)

but, referring to the entropy inequality (9.64) which can be cast in the form

bγq

2

≥ 0, it follows that b < 0 whence L ≡ (bγ)

−1

≥ 0 and τ = −(ρ/b) ≥ 0.

These results conﬁrm the positive deﬁnite property of the heat conductivity

coeﬃcient and the relaxation time.

As compared with the results of Appendix 1, it is seen that in the present

description, the non-equilibrium entropy (9.73) depends on the heat ﬂux. This

is due to the fact that q is assumed to have an evolution equation (9.58) of

its own, leading to the introduction of a supplementary Lagrange multiplier

in equation (9.60); instead, in Appendix 1 the temperature gradient was not

assumed to be described by a proper evolution equation so that no extra

Lagrange multiplier was needed.

9.6 Problems

9.1. Dissipated energy. Verify that the left-hand side of the Clausius–Duhem’s

inequality (9.5) can be identiﬁed with the rate of dissipated energy Tσ

s

at

the condition that the corresponding entropy ﬂux is given by J

s

= q/T.

9.2. Generalized Fourier’s law. Show that the generalized Fourier law as the

time derivative of the heat ﬂux q(t) =

_

t

−∞

λ(t −t

)∇T(t

)dt

is equal to

Cattaneo’s relation ˙ q = −q/τ − (λ/τ)∇T when the memory kernel is given

by the expression λ(t −t

) = −(λ/τ) exp[−(t −t

**)/τ], where λ is the heat
**

conductivity and τ is a constant relaxation time.

9.3. Objectivity. Show that the skew-symmetric (antisymmetric) part of the

velocity gradient tensor W =

1

2

[(∇v) −(∇v)

T

] transforms as W

∗

= Q W

Q

T

+

˙

Q Q

T

under the Euclidean transformation (9.8).

258 9 Rational Thermodynamics

9.4. Objectivity. Prove that (a) ∇T and the displacement vector u of elas-

ticity are objective vectors; (b) the symmetric part of the velocity gradient

V =

1

2

[(∇v) + (∇v)

T

] and the gradient of an objective vector are objective

tensors; and (c) the deformation gradient F and the material time derivative

of an objective vector, say the heat ﬂux q, are not objective.

9.5. Clausius–Duhem’s inequality. Establish the Clausius–Duhem’s inequal-

ity, respectively, in the case of small elastic deformations (see (9.18)) and

large deformations (refer to Box 9.1).

9.6. Large elastic deformations. Consider a material body deﬁned by the

set of variables T, ∇T, F, and

˙

F where T is the temperature and F the

deformation tensor. Determine the restrictions placed on the constitutive

equations of f (free energy), s (entropy), S (Piola stress tensor), and Q

(heat ﬂux vector).

9.7. Isotropic tensors. Referring to the theorems of representation of isotropic

tensors (e.g. Truesdell and Toupin 1960), the constitutive equation of the

viscous stress tensor of a heat conducting viscous ﬂuid reads as

σ

v

= α(ρ, T)I +γ(ρ, T)(∇ v)I + 2η(ρ, T)V,

when the second- and higher-order terms in V are omitted. (a) Show that α

is zero, as a result of the property that the viscous stress tensor has to be

zero at equilibrium. (b) Decomposing the symmetric velocity gradient tensor

in a bulk and a deviatoric part V =

1

3

(∇ v)I +

0

V, with

0

V the traceless

deviator, show that

σ

v

= ζ(ρ, T)(∇ v)I + 2η(ρ, T)

0

V,

with ζ = γ +

2

3

η designating the bulk viscosity. (c) Verify that ζ > 0, η > 0.

9.8. Parabolic heat conduction. Consider a one-dimensional rigid heat con-

ductor deﬁned by the following constitutive equations:

f = f(T, ∂T/∂x), s = s(T, ∂T/∂x), q = q(T, ∂T/∂x).

and the energy balance equation

ρ∂u/∂t = ∂q/∂x +ρr.

Determine the resulting heat conduction equation for the temperature ﬁeld,

given by

ρT

∂

2

f

∂T

2

∂T

∂t

= −

∂q

∂T

∂T

∂x

−

∂q

∂(∂T/∂x)

∂

2

T

∂x

2

+ρr.

Show that this equation is of the parabolic type if ∂

2

f/∂T

2

> 0 and ∂q/∂T <

0 so that no propagation of wave can occur.

9.6 Problems 259

9.9. Hyperbolic heat conduction. Reconsider Problem 9.8 but with T, ∂T/∂t,

∂T/∂x as variables. Show that in the present case, the temperature equation

is of the hyperbolic type.

9.10. Viscous heat conducting ﬂuid. A viscous heat conducting ﬂuid is char-

acterized by the following constitutive equation φ = φ(v, T, ∇v, ∇T, V) with

φ ≡ f, s, q, σ. Determine the restrictions placed by Clausius–Duhem’s in-

equality on the constitutive equations.

9.11. Thermodiﬀusion. Study the problem of thermodiﬀusion in a binary

mixture of non-viscous ﬂuids within the framework of rational thermody-

namics.

9.12. Liu’s technique. Reformulate the problem of a heat conducting viscous

ﬂuid (see Sect. 9.4) by using Liu’s technique.

Chapter 10

Hamiltonian Formalisms

A Mathematical Structure of Reversible

and Irreversible Dynamics

In the theories discussed so far, the limitations on the form of the constitutive

equations arise essentially from the application of the second law of thermo-

dynamics. However, no restrictions were placed on the reversible terms of the

time evolution equations, as they do not contribute to the entropy production.

Such terms may be either gyroscopic forces, as Coriolis force, or convected

time derivatives, as Maxwell or corotational derivative, which are of frequent

use in rheology. It is shown here that requirement of a Hamiltonian structure

leads to restrictions on the reversible part of the evolution equations.

Hamiltonian formulations have always played a central role not only in

mechanics but also in thermodynamics. They have been identiﬁed at dif-

ferent levels of description: from the microscopic one (classical mechanics,

kinetic theory) to the macroscopic one (theory of elasticity, frictionless ﬂuid

ﬂows, equilibrium thermodynamics). Despite their theoretical appeal, and

except rare attempts, Hamiltonian methods have not been fully exploited

in presence of irreversible eﬀects. The reasons that militate in favour of a

Hamiltonian description are numerous: the ﬁrst one is conciseness, as the

whole set of balance equations are expressed in terms of a limited number of

potentials, generally, one single potential is suﬃcient. From a practical point

of view, there exist, in parallel, powerful numerical methods, which have

been developed for Hamiltonian systems to obtain approximate solutions, as

the task of scientists is not only to derive equations but also to ﬁnd solu-

tions. Moreover, besides their power of synthesis and their practical interest,

Hamiltonian descriptions are also helpful for the physical interpretation of

the processes: indeed, the generating potentials may generally be identiﬁed

with well-deﬁned physical quantities as the mechanical energy, the energy

of deformation, the thermodynamic potentials, etc. Hamiltonian techniques

have also a wide domain of applicability, as they are not restricted to the

linear regime. Finally, they place several restrictions on the possible forms of

the constitutive equations, complementing those provided by the second law

of thermodynamics.

261

262 10 Hamiltonian Formalisms

The main part of this chapter will be devoted to the presentation of a

general equation for the non-equilibrium reversible–irreversible coupling, ab-

breviated as GENERIC, which consists in a generalization of the Poisson

bracket formalism originally proposed in the framework of classical mechan-

ics (Grmela and

¨

Ottinger 1997;

¨

Ottinger and Grmela 1997;

¨

Ottinger 2005).

The principal motivation for developing GENERIC is the modelling of the

ﬂow properties of rheological ﬂuids; this is achieved by formulating general

time evolution equations taking the same universal form whatever the nature

of the state variables. Another particularity of GENERIC is that the evo-

lution equations are expressed in terms of two appropriate thermodynamic

potentials, called generators, taking the form of the total energy and a dissi-

pation potential. GENERIC can be viewed as an extension of both Hamilton’s

equations and Landau’s potential (Landau 1965). Before examining the main

tenets of GENERIC and illustrating its application by a number of simple

examples, let us preliminarily recall the Hamiltonian description of classical

mechanics.

10.1 Classical Mechanics

Consider a collection of N particles characterized by the set of variables

x = (r, p) with r = (r

1

, . . . , r

N

) and p = (p

1

, . . . , p

N

) denoting the position

and momentum vectors of the particles. Denoting by E = K + V the total

energy (kinetic and potential energies) or the Hamiltonian of the system, it

is well known that the time evolution of the set of variables x is given by the

Hamilton equations

dx

dt

=

ˆ

L

∂E

∂x

, (10.1)

with

ˆ

L =

_

0 1

−1 0

_

,

or, more explicitly,

dr

dt

=

∂E

∂p

,

dp

dt

= −

∂E

∂r

. (10.2)

The Poisson matrix

ˆ

L expresses the reversible kinematics of x; reversibility

means invariance with respect to the change t →−t, it being understood that

concomitantly r, p → r, −p and E(r, p) → E(r, −p). Instead of working

with the matrix

ˆ

L, an equivalent way to formulate the dynamical equations

(10.2) is to introduce the Poisson bracket

¦A, B¦ =

∂A

∂r

∂B

∂p

−

∂A

∂p

∂B

∂r

, (10.3)

10.1 Classical Mechanics 263

which is related to

ˆ

L by

¦A, B¦ =

_

∂A

∂x

,

ˆ

L

∂B

∂x

_

(10.4)

with (,) denoting the scalar product, whereas A and B are regular functions

of x. The evolution of x is now governed by

dA

dt

= ¦A, E¦, (10.5)

where A(x) is an arbitrary function of x, not dependent explicitly on time.

This is directly seen by writing explicitly both members of (10.5), i.e.

∂A

∂r

dr

dt

+

∂A

∂p

dp

dt

=

∂A

∂r

∂E

∂p

−

∂A

∂p

∂E

∂r

, (10.6)

and after identifying the coeﬃcients of ∂A/∂r and ∂A/∂p in both sides of

(10.6), one recovers indeed the evolution equations (10.2). An illustrative

example is presented in Box 10.1.

Box 10.1 A Classical Mechanics Illustration: The Harmonic

Oscillator

Consider a particle of constant mass m ﬁxed at the end of a spring of

stiﬀness H. The problem is assumed to be one dimensional with x denoting

the position of the particle and p its momentum. The total energy is

E(x, p) =

1

2

p

2

m

+

1

2

Hx

2

. (10.1.1)

In virtue of (10.2), the evolution equations of x and p are given by

dx

dt

=

p

m

,

dp

dt

= −Hx. (10.1.2)

After elimination of p, one recovers Newton’s law

m

d

2

x

dt

2

= −Hx. (10.1.3)

Making use of the deﬁnition (10.4) of the Poisson bracket ¦A, B¦ one has

¦A, B¦ =

_

∂A

∂x

∂A

∂p

__

0 1

−1 0

_

⎛

⎜

⎜

⎝

∂B

∂x

∂B

∂p

⎞

⎟

⎟

⎠

, (10.1.4)

which is in agreement with the result (10.3). Accordingly, the Poisson

bracket ¦A, E¦ describing the harmonic oscillator is given by

¦A, E¦ =

∂A

∂x

p

m

−

∂A

∂p

Hx (10.1.5)

264 10 Hamiltonian Formalisms

and the corresponding evolution equation (10.5) takes the form

∂A

∂x

dx

dt

+

∂A

∂p

dp

dt

=

∂A

∂x

p

m

−

∂A

∂p

Hx. (10.1.6)

Comparison of the coeﬃcients of ∂A/∂x and ∂A/∂p in both sides of (10.1.6)

gives back the equations of motion (10.1.2).

Expression (10.4) is a Poisson bracket if the two following conditions are

fulﬁlled:

(1)

¦A, B¦ = −¦B, A¦ (antisymmetry), (10.7)

(2)

¦A, ¦B, C¦¦ +¦B, ¦C, A¦ +¦C, ¦A, B¦¦ = 0 ( Jacobi’s identity). (10.8)

The antisymmetry property of the Poisson bracket implies the antisymme-

try of operator

ˆ

L; the Jacobi’s identity imposes additional severe restrictions

and expresses the time structure invariance of the Poisson bracket ¦A, B¦.

It is precisely this identity what gives information on the reversible contri-

butions to the dynamics (for instance, non-linear convective contributions

to the time derivatives), which is not available from the positiveness of the

entropy production.

Evolution equations as (10.1) or its equivalent (10.5) are reversible time

evolution equations as they are invariant with respect to time reversal t →−t,

a result well known in classical mechanics. In more general situations as

these encountered in continuum physics, dx/dt is the sum of a reversible

and a non-reversible contribution and we will see in Sect. 10.2 how to inte-

grate irreversible dynamics in the framework of GENERIC. In mathematical

terminology, the manifold of the phase space, a Hamiltonian, and an antisym-

metric matrix relating the time derivative of the variables with the partial

derivatives of the Hamiltonian with respect to its variables are known as a

symplectic manifold.

10.2 Formulation of GENERIC

GENERIC may be considered as an extension of Landau’s idea accordingly

the time evolution of a state variable x, like mass density or energy, towards

its equilibrium value x

eq

is governed by the relation

dx

dt

= −

δΨ

δx

, (10.9)

10.2 Formulation of GENERIC 265

where Ψ is a given potential which is minimum in the equilibrium or in the

steady state:

δΨ

δx

= 0 at x = x

eq

, (10.10)

with δ/δx denoting the functional or Volterra derivative with respect to the

variable x; if Ψ =

_

ψ dV is a simple scalar function of the variables, then

δΨ/δx reduces to the usual partial derivative δΨ/δx = ∂ψ/∂x; the situation

is more complicated when Ψ depends in addition on the gradient of x, if x

is assumed to be a scalar, then

δΨ

δx

=

∂ψ

∂x

−∇

_

∂Ψ

∂(∇x)

_

. (10.11)

However, a relaxation-type equation as (10.9) describing the irreversible ap-

proach to equilibrium is too restrictive and cannot pretend to cope with

general processes of continuum physics.

When both reversible and irreversible processes are present, one formu-

lates, in the framework of GENERIC, a general time evolution equation in

which the evolution of a variable x is expressed in terms of two potentials, the

total energy E of the overall system and a dissipation function Ψ: explicitly,

one has

dx

dt

=

ˆ

L

δE

δx

+

δΨ

δ(δS/δx)

. (10.12)

The quantity S has the physical meaning of the entropy of the overall system.

The dissipation potential Ψ, which is a real-valued function of the derivatives

δS/δx, possesses the following properties: Ψ(0) = 0, it is minimum at x =0

and is convex in the neighbourhood of 0. Note that in the GENERIC frame-

work, it is generally assumed that the overall system is isolated from its

environment so that

dE

dt

= 0,

dS

dt

> 0. (10.13)

Notice also that in the particular case that Ψ is a quadratic function of δS/δx,

of the form

1

2

(∂S/∂x)

ˆ

M (∂S/∂x), (10.12) takes the more familiar form

dx

dt

=

ˆ

L

δE

δx

+

ˆ

M

δS

δx

. (10.14)

The matrices

ˆ

L and

ˆ

M, operating on the functional derivatives of E and S,

produce the reversible and irreversible contributions to the evolution of x,

respectively. These matrices must satisfy some conditions:

ˆ

L must be an-

tisymmetric and verify Jacobi’s identity, and

ˆ

M must be symmetric and

positive-deﬁnite to ensure the positiveness of the dissipation rate. However,

the restrictions on

ˆ

L and

ˆ

M mentioned above are not suﬃcient to guarantee a

thermodynamically consistent description of the dynamics of the system, and

266 10 Hamiltonian Formalisms

two supplementary restrictions, called degeneracy conditions, are introduced,

namely

ˆ

L

δS

δx

= 0, (10.15)

ˆ

M

δE

δx

= 0. (10.16)

The ﬁrst condition expresses that entropy cannot contribute to the reversible

nature of

ˆ

L and therefore is not modiﬁed by the reversible part of the dy-

namics; the second condition implies the conservation of total energy by the

dissipative contribution to the dynamics. Relations (10.12) or (10.14) repre-

sent the GENERIC extension of the Landau equation (10.9) and express the

universal structure of non-equilibrium thermodynamics, which is completely

speciﬁed by the knowledge of the four quantities E, S (or Ψ),

ˆ

L, and

ˆ

M.

By analogy with classical mechanics where the Hamilton equations (10.1)

can be replaced by the Poisson brackets (10.5), it is equivalent to write (10.14)

in the form

dA

dt

= ¦A, E¦ + [A, S], (10.17)

where

¦A, E¦ =

_

δA

δx

,

ˆ

L

δE

δx

_

(10.18)

is a Poisson bracket with the antisymmetry property ¦A, E¦ = −¦E, A¦,

whereas

[A, S] =

_

δA

δx

,

ˆ

M

δS

δx

_

(10.19)

is the so-called Landau symmetric bracket, i.e. [A, S] = [S, A] satisfying in

addition the positiveness property [S, S] > 0. In terms of the dissipation

potential Ψ, the above bracket will be given the form

[A, S] =

_

δA

δx

,

δΨ

δ(δS/δx)

_

, (10.20)

which is a generalization of (10.19). In the foregoing, we will illustrate the

use of GENERIC with two examples: isothermal hydrodynamics and matter

diﬀusion in a binary moving mixture. Several applications to polymer solu-

tions have been worked out, but we refer to

¨

Ottinger and Grmela (1997) and

¨

Ottinger (2005) for an overview of this topic.

10.2.1 Classical Navier–Stokes’ Hydrodynamics

Let us consider the motion of a compressible one-component viscous ﬂuid

assumed to take place under isothermal conditions. The following four steps

govern the construction of the GENERIC formalism:

10.2 Formulation of GENERIC 267

Step 1. Selection of the state variables

Just like in other thermodynamic theories, the selection of state variables is

subordinated by the nature of the process under consideration and the degree

of accuracy that one wishes to achieve. For the present problem, the set x of

variables is conveniently selected as:

x : ρ(mass density), u(momentum). (10.21)

In GENERIC, it is more convenient to work with the momentum rather than

with the velocity ﬁeld v(= u/ρ). The pressure p is not included among the

variables because it will be expressed in terms of the independent ﬁelds by

means of a constitutive equation.

Step 2. Thermodynamic potential

Since the temperature T is ﬁxed and uniform, the Helmholtz free energy is

the potential that will play the central role, it is expressed by

Φ(ρ, u) =

_

φ(ρ, u)dV = E(ρ, u) −TS(ρ), (10.22)

with the total energy and the total entropy given by

E(ρ, u) =

_

[(u u)(2ρ)

−1

+ε(ρ)]dV, (10.23a)

S(ρ) =

_

s(ρ)dV , (10.23b)

where ε and s denote, respectively, the internal energy and the entropy re-

ferred per unit volume; just like ε, the entropy s cannot depend on the

momentum u as it is a speciﬁc thermodynamic quantity. Here we use the

notation ε instead of u to designate the internal energy, thus avoiding the

confusion with the momentum u.

For further purpose, it is interesting to observe that φ

u

α

= u

α

/ρ is the

velocity ﬁeld; the notation φ

u

α

standing for ∂φ/∂u

α

where ∂/∂u

α

is the

partial derivative with respect to the component u

α

of the momentum.

Step 3. Hamiltonian reversible dynamics

Our purpose is to specify the reversible contribution to the evolution equa-

tions of hydrodynamics. Let A =

_

a dV denote a regular function of ρ and

u whose time evolution is given by

dA

dt

=

_

(a

ρ

∂

t

ρ +a

u

α

∂

t

u

α

)dV, (10.24)

where a

ρ

= ∂a/∂ρ and a

u

α

= ∂a/∂u

α

.

Following the lines of thought of GENERIC, the time evolution of A can

be cast in the form

dA

dt

= ¦A, Φ¦, (10.25)

268 10 Hamiltonian Formalisms

where ¦A, Φ¦ is a Poisson bracket which, in the present problem, is deﬁned as

¦A, Φ¦ =

_

¦ρ[(∂

α

a

ρ

)φ

u

α

−(∂

α

φ

ρ

)a

u

α

] +u

γ

[(∂

α

a

u

γ

)φ

u

γ

−(∂

α

φ

u

γ

)a

u

α

]¦dV

(10.26)

with ∂

α

denoting the spatial derivative with respect to the x

α

coordinate,

we have also used the summation convention on repeated indices. A system-

atic construction of the expression of the Poisson bracket (10.26) has been

developed on general arguments based on the group theory by Grmela and

¨

Ottinger (1997) and

¨

Ottinger (2005). Substituting (10.24) and (10.26) in

(10.25), one obtains

_

[a

ρ

(∂

t

ρ) +a

u

α

(∂

t

u

α

)]dV

=

_

_

a

ρ

[−∂

γ

(ρφ

u

γ

)] −a

u

α

[∂

γ

(u

α

φ

u

γ

) +ρ∂

α

φ

ρ

+u

γ

∂

α

u

γ

]

_

dV, (10.27)

after that the right-hand side of (10.26) has been integrated by parts and

the boundary conditions have been selected to make all the integrals over the

boundary equal to zero.

By identiﬁcation of the coeﬃcients of a

ρ

and a

u

α

, it is found that

∂

t

ρ = −∂

γ

(ρφ

u

γ

), (10.28)

∂

t

u

α

= −∂

γ

(u

α

φ

u

γ

) −ρ∂

α

φ

ρ

−u

γ

∂

α

φ

u

γ

. (10.29)

To recover the Euler equations of hydrodynamics, one has to identify the two

last terms of the right-hand side of (10.29) with the pressure gradient, i.e.

∂

α

p = ρ∂

α

φ

ρ

+u

γ

∂

α

φ

u

γ

. (10.30)

Moreover, from the chain diﬀerentiation rule, one obtains

∂

α

φ = φ

ρ

∂

α

ρ +φ

u

γ

∂

α

u

γ

,

which, coupled to (10.30) leads to

p = −φ +ρφ

ρ

+u

γ

φ

u

γ

= −ε +Ts +ρµ. (10.31)

This is the well-known Euler equation of equilibrium thermodynamics after

that φ

ρ

= µ has been identiﬁed with the chemical potential. As shown earlier,

one has φ

u

α

= v

α

so that (10.28) and (10.29) take the form

∂

t

ρ = −∂

γ

(ρv

γ

), (10.32)

∂

t

u

α

= −∂

γ

(pδ

αγ

+u

α

u

γ

ρ

−1

), (10.33)

which are the equations describing the behaviour of an Euler ﬂuid in hydro-

dynamics.

10.2 Formulation of GENERIC 269

When the time evolution of the system is described by the Euler equations

(10.32) and (10.33), it is checked that

dΦ

dt

= 0, (10.34)

which follows directly from (10.25) as a consequence of the antisymmetry

property of the Poisson bracket. It is also veriﬁed that the relations (10.32)

and (10.33) are invariant with respect to the time reversal t → −t and this

justiﬁes the denomination “reversible” or “non-dissipative”.

Step 4. Irreversible dynamics

In the presence of dissipation, the free energy Φ will diminish in the course

of time

dΦ

dt

< 0 (10.35)

and the general time evolution equation is

dA

dt

= ¦A, Φ¦ −[A, Φ], (10.36)

where, referring to (10.20), the dissipative bracket is deﬁned as

[A, Φ] = [δA/δu

α

, δΨ/δφ

u

α

] =

_

a

u

α

[(∂ψ/∂φ

u

α

) −∂

γ

(∂ψ/∂(∂

γ

φ

u

α

))] dV

(10.37)

with Ψ =

_

ψ dV the dissipation potential. Two remarks are in form about

the formulation of (10.36) and (10.37). First, when compared to (10.17), the

change of sign in front of the dissipative bracket is a consequence of the use

of the free energy Φ instead of the entropy S as basic function; second, there

is no contribution of the mass density to (10.37) because ρ is a conserved

quantity.

To reproduce the classical Navier–Stokes’ relation, select ψ as given by the

quadratic form

ψ =

1

4

(∂

γ

φ

u

a

+∂

α

φ

u

γ

)η(∂

γ

φ

u

α

+∂

α

φ

u

γ

) > 0, (10.38)

wherein η is a phenomenological coeﬃcient, the dynamic viscosity, which is a

positive quantity to guarantee the positive deﬁnite property of the dissipation

potential. It is directly checked that the integrant of (10.37) is

−a

u

α

∂

γ

_

η(∂

α

φ

u

γ

+∂

γ

φ

u

α

)

¸

. (10.39)

Following the same procedure as in Step 3, one obtains the next evolution

equations for ρ and u

α

:

∂

t

ρ = −∂

γ

(ρφ

u

γ

), (10.40)

∂

t

u

α

= −∂

γ

(u

α

φ

u

α

) −∂

α

p −∂

γ

P

αγ

, (10.41)

270 10 Hamiltonian Formalisms

where the symmetric pressure tensor P

αγ

is expressed by

P

αγ

= −

1

2

η(∂

α

φ

u

γ

+∂

γ

φ

u

α

) = −

1

2

η(∂

α

v

γ

+∂

γ

v

α

). (10.42)

The last term in (10.41) represents the irreversible contribution to the mo-

mentum equation as it is not invariant to the time reversal. From the iden-

tiﬁcation of (10.42), it is seen that the dissipation potential (10.38) can be

cast in the form −P

v

: V, which is (2.68) for the viscous dissipation found

in the classical formulation of hydrodynamics. By incorporating in the rela-

tion (10.38) of the dissipation potential an extra term proportional to the

square of ∂

γ

φ

u

γ

, one would recover Stokes’ law between the bulk viscous

pressure and the divergence of the velocity. External forces, like gravity, can

also be incorporated by simply adding the corresponding potential, namely

−

_

ρg x dV to the total energy given by (10.23a).

The above results are important because they exhibit the GENERIC struc-

ture of the familiar Navier–Stokes’ equations. Indeed, accordingly to the gen-

eral rules of GENERIC, the evolution equations of hydrodynamics can be

formulated in terms of two potentials: the Helmholtz free energy and the

dissipation potential; moreover, there exists an acute separation between re-

versible and irreversible contributions. It is also important to mention some

properties of the solutions of (10.40) and (10.41): ﬁrst, when the system is left

during a suﬃciently long time outside the inﬂuence of external constraints, it

will tend to an equilibrium state characterized by φ

u

α

= 0, φ

ρ

= 0 for t →∞.

Second, if the potential Φ is convex (Φ ≥ 0), and because dΦ/dt < 0, Φ plays

the role of a Lyapounov function ensuring the stability of equilibrium.

It was stated in Sect. 10.1 that an equivalent description of GENERIC

consists in expressing the evolution equations in terms of the two potentials

E and S and the operators

ˆ

L and

ˆ

M. However, this kind of approach re-

quires some expertise, technique and feeling, which is outside the scope of

the present introductory monograph and therefore we refer the reader to

specialized works (Grmela and

¨

Ottinger 1997;

¨

Ottinger and Grmela 1997;

¨

Ottinger 2005).

In the present section, we have considered the ﬂow of a one-component

ﬂuid. In Sect. 10.2.2, we will examine the more general situation of the motion

of a binary mixture to exhibit the coupling between matter diﬀusion and

momentum transport.

10.2.2 Fickian Diﬀusion in Binary Mixtures

We consider an isotropic mixture of two Newtonian viscous ﬂuids perfectly

miscible and chemically inert, the whole system is in motion in absence of

external forces and its temperature is assumed to be uniform. Let ρ

1

and ρ

2

be the mass densities of the two components and u

1

and u

2

their barycentric

10.2 Formulation of GENERIC 271

momenta, the average barycentric momentum is given by u = u

1

+u

2

. The

presentation of Sect. 10.2.1 will be traced back and ﬁrst applied to Fickian

diﬀusion.

Step 1. State variables

The space of basic variables for the present problem is formed by the set

x : u, ρ

1

, ρ

2

. However, for practical reasons, it is more convenient to work

with the following equivalent set

x : u, ρ, c, (10.43)

where ρ = ρ

1

+ρ

2

is the total mass density and c = ρ

1

/ρ is the mass fraction

of one of the constituents, say component 1 (note that c + c

2

= 1, with

c

2

= ρ

2

/ρ).

Step 2. Thermodynamic potential

We will use the same Helmholtz potential as in Sect. 10.2.1, i.e.

Φ(ρ, c, u) = E(ρ, c, u) −TS(ρ, c) (10.44)

with E =

_

[u u/(2ρ) +ε(ρ, c)]dV where ε is the total internal energy per

unit volume of the binary mixture. At this level of description, we have ne-

glected the contribution from the diﬀusion kinetic energy, which is justiﬁed

in diluted systems.

Step 3. Reversible dynamics

The reversible dynamics is written in the form dA/dt = ¦A, Φ¦ where the

Poisson bracket is now given by (Grmela et al. 1998):

¦A, Φ¦ =

_

_

ρ[(∂

α

a

ρ

)φ

u

α

−(∂

α

φ

ρ

)a

u

α

] +u

γ

[(∂

α

a

u

γ

)φ

u

α

−(∂

α

φ

u

γ

)a

u

α

] −(∂

α

c)(a

c

φ

u

α

−φ

c

a

u

α

)

_

dV. (10.45)

This is the same relation as (10.26) to which we have added a third-term

expressing the contribution from the mass fraction c. The resulting evolution

equations are obtained as before, i.e. by writing explicitly dA/dt in terms of

the variables, by integrating (10.45) by parts and by identifying the coeﬃ-

cients of a

ρ

, a

u

α

, and a

c

, respectively. These operations lead to the following

reversible part of the evolution equations

∂

t

ρ = −∂

γ

(ρφ

u

γ

), (10.46)

∂

t

u

α

= −∂

γ

(u

α

φ

u

γ

) −∂

α

p, (10.47)

∂

t

c = −(∂

γ

c)φ

u

γ

, (10.48)

where the thermostatic pressure is still given by p = −φ + ρφ

ρ

+ u

γ

φ

u

γ

. It

is easily checked that the above equations are invariant with respect to time

reversal.

272 10 Hamiltonian Formalisms

Step 4. Irreversible dynamics

To recover Fick’s law of diﬀusion, it is necessary to introduce the dissipa-

tion potential. The latter is a generalization of the potential used in (10.38),

written here as ψ (10.38), with a supplementary term,

ψ = ψ(10.38) +

1

2

(∂

γ

φ

c

)D

∗

(∂

γ

φ

c

). (10.49)

The contribution coming from diﬀusion is the last term at the right-hand side

of (10.49), D

∗

is a positive phenomenological coeﬃcient allowed to depend

on ρ and c. After that dissipation has been included, the evolution equations

(10.46)–(10.48) take the ﬁnal form

∂

t

ρ = −∂

γ

(ρv

γ

), (10.50)

∂

t

u

α

= −∂

γ

(u

α

v

γ

) −∂

α

p −∂

γ

P

αγ

, (10.51)

∂

t

c = −v

γ

∂

γ

c +∂

γ

(D

∗

∂

γ

φ

c

), (10.52)

wherein the two ﬁrst relations are the classical continuity and momentum

equations whereas the last one is the diﬀusion equation; the irreversible con-

tributions are the last term in (10.51), with P

αγ

still given by (10.42), and the

last term of (10.52). Denoting by J

γ

, the γ component of the ﬂux of matter,

it follows from (10.52) that

J

γ

= −D

∗

∂

γ

φ

c

. (10.53)

Assuming that φ is a quadratic function of c with φ

c

= D

1

c, where D

1

is a

phenomenological constant, one recovers Fick’s law

J

γ

= −ρD∂

γ

c, (10.54)

at the condition to deﬁne the coeﬃcient of diﬀusion D as D

1

D

∗

/ρ; observe

that in contrast to D

1

, the factor D is not necessarily a constant.

One of the merits of GENERIC is to provide a systematic way to derive

the evolution equations whatever the nature and the number of basic vari-

ables. Of course, to obtain explicit relations, we need constitutive relations

for the potentials φ and ψ, which will be determined by the physico-chemical

properties of the system under study.

Although the above analysis gives an overall description of diﬀusion in

a two-constituent mixture, it cannot pretend to describe the behaviour of

each individual component; in particular, one has no information about the

time evolution of the particular velocities u

1

and u

2

. This would require the

introduction of u

1

and u

2

as independent variables, instead of the single

barycentric momentum u. Such a task will be achieved in Sect. 10.2.3.

10.2 Formulation of GENERIC 273

10.2.3 Non-Fickian Diﬀusion in Binary Mixtures

Our objective is twofold: to derive the evolution equations of the two indi-

vidual components, and to generalize the Fick’s law to non-steady situations.

For the sake of simplicity, we assume that the barycentric momentum u is

equal to zero; this assumption implies that the total variation of the total

mass density ρ is negligible in the course of time. In addition, we assume

that both ﬂuids are non-viscous. A more general analysis, wherein the overall

velocity does not vanish and where viscosity is not neglected, can be found

in Grmela et al. (2003).

1. State variables

Here we make the one-to-one transformation

ρ

1

, ρ

2

, u

1

, u

2

→ρ, c, u, w, (10.55)

where as above ρ designates the total mass density, c is the mass fraction

of component 1, u is the barycentric momentum while the vector w stands

for

w = ρ

−1

(ρ

2

u

1

−ρ

1

u

2

). (10.56)

As stated above, from now on, one supposes that u = 0 (no convection).

Once c, ρ, and w are determined, one obtains directly ρ

1

, ρ

2

, u

1

and u

2

from the inverse transformation ρ

1

= cρ, ρ

2

= (1 −c)ρ, u

1

= −u

2

= w.

2. Thermodynamic potential

The latter is similar to (10.44) but the speciﬁc kinetic energy u

2

1

(2ρ

1

)

−1

+

u

2

2

(2ρ

2

)

−1

is now equal to the diﬀusion kinetic energy w

2

/2ρc(1−c) instead

of the barycentric kinetic energy u

2

/2ρ, as in Sect. 10.2.2.

3. Hamiltonian reversible dynamics

The Hamiltonian setting is equivalent to that proposed in the previous

sections; inspired by (10.26) and (10.45), we will write

¦A, Φ¦ =

_

¦ρc(1 −c)[∂

α

(a

c

/ρ)φ

w

α

−∂

α

(φ

c

/ρ)a

w

α

]

−(1 −c)w

γ

[(∂

α

a

w

γ

)φ

w

α

−(∂

α

φ

w

γ

)a

w

α

]

−w

γ

[∂

α

(ca

w

γ

)φ

w

α

−∂

α

(cφ

w

γ

)a

w

α

]

_

dV. (10.57)

It follows that the non-dissipative contribution to the evolution equations

will be

∂

t

ρ = 0, (10.58)

∂

t

c = −∂

γ

[ρc(1 −c)φ

w

γ

]/ρ, (10.59)

∂

t

w

α

= −ρc(1 −c)∂

α

(φ

c

/ρ) +c∂

γ

(w

α

φ

w

γ

)

−∂

γ

[(1 −c)w

α

φ

w

γ

] −(1 −c)w

γ

(∂

α

φ

w

γ

) +w

γ

∂

α

(cφ

w

γ

). (10.60)

Expression (10.59) is the classical diﬀusion equation wherein the ﬂux of

matter is given by J

γ

= ρc(1 − c)φ

w

γ

. Relation (10.60) represents the

274 10 Hamiltonian Formalisms

most original contribution of GENERIC formalism to this problem, as

it is a time evolution equation for the relative velocity w, which has no

counterpart in the classical theory of irreversible processes. More about its

physical content will be discussed below.

4. Dissipation

As viscous contributions are neglected, the dissipation potential is simply

given by

ψ =

1

2

φ

w

α

Λφ

w

α

, (10.61)

where Λ is a phenomenological coeﬃcient which is positive to guarantee

the positiveness of the dissipation potential. The complete set of evolution

equations is still given by (10.58)–(10.60) at the condition to complement

the evolution equation (10.60) for w

α

by the dissipative term −Λφ

w

α

.

It is interesting to observe that by inserting this dissipative term in (10.60),

one obtains an equation of the Cattaneo type. By omitting in (10.60) the

non-linear contributions in the ﬂuxes and making use of the result φ

w

α

=

w

α

/ρc(1 −c), one arrives at

∂

t

w

α

+ [Λ/ρc(1 −c)]w

α

= −ρc(1 −c)∂

α

(φ

c

/ρ), (10.62)

which is clearly of the form

τ∂

t

w

α

+w

α

= −D

∗

∂

α

(φ

c

/ρ), (10.63)

where τ = 2ρc(1 −c)Λ

−1

corresponds to the relaxation time of the diﬀusive

momentum and D

∗

= [ρc(1 − c)]

2

Λ

−1

to a diﬀusion-like coeﬃcient. Expres-

sion (10.63) is an example of non-Fickian law. Non-Fickian diﬀusion occurs for

instance when a solvent diﬀuses into a medium characterized by a timescale

which is the same or larger than the timescale on which the penetration

takes place. This situation arises when solvents penetrate into glassy poly-

mers. Another example in which diﬀusion exhibits non-classical features is the

mixture of

3

He–

4

He isotopes below the lambda point (Lhuillier et al. 2003;

Lebon et al. 2003).

If we assume that w

α

evolves much faster than the other state variables

(∂

t

w

α

=0), then (10.63) reduces to the standard description of Fick’s dif-

fusion. Indeed, in virtue of the deﬁnition of the diﬀusion ﬂux J

α

= ρc(1 −

c)φ

w

α

≡ w

α

, (10.63) becomes simply

J

α

= −D

∗

∂

α

(φ

c

/ρ), (10.64)

which is the classical Fick’s law.

10.3 Final Comments

In this section, GENERIC has been presented as a general formalism to de-

scribe systems driven out of equilibrium. The main feature of this theory is

10.3 Final Comments 275

that it uses two generating potentials: the total energy for reversible dynamics

and entropy (or more generally the dissipation potential) for irreversible dy-

namics. The principal objective is to derive the relevant evolution equations

of the process under study. The main result of the study is that it oﬀers a

structure which is applicable whatever the level of description, either macro-

scopic, mesoscopic, or microscopic. The basic ideas underlying GENERIC

have been discussed thanks to two simple examples: classical hydrodynamics

and diﬀusion of two perfectly miscible ﬂuids.

It is instructive to recapitulate how we obtained the time evolution equa-

tions. First, one has to deﬁne the space of the state variables x. This choice

depends on our goals and the nature of the system under investigation; it

is made on the basis of some intuition guided by our insight of the physics

of the problem. In classical hydrodynamics, the solution is unambiguous:

the state variables are the classical hydrodynamic ﬁelds, i.e. mass, momen-

tum, and energy or temperature. However, this classical approach fails when

dealing with complex ﬂuids, like non-Newtonian or polymer solutions. Due

to the active role of the microscopic structure of the complex ﬂuids in the

macroscopic time evolution, the choice of the state variables is not universal.

For example, to describe complex ﬂuids, the hydrodynamic variables will be

complemented by structural variables, sometimes called internal variables as

in Chap. 8. This is in opposition with the use of functionals as in rational

thermodynamics. We emphasize that the choice of state variables represents

always a ﬁrst step to make and that this is always an assumption. It should

also be stressed that GENERIC does not oﬀer any help in the selection of

the basic variables.

The next step is to formulate the corresponding set of evolution equations.

In classical hydrodynamics, they are the usual balance equations of mass,

momentum, and energy complemented by appropriate constitutive equations

allowing us to “close” the set of balance equations. In GENERIC, one makes

a clear distinction between the reversible and the irreversible contributions

to the kinematics of the ﬁeld variables. The non-dissipative part is obtained

explicitly as a straightforward consequence of the dynamics expressed by

the Poisson brackets or equivalently by the skew-symmetric operator

ˆ

L. The

reversible part of the evolution equation is completely speciﬁed by the kine-

matics, which means that explicit relations are obtained for the evolution

equations including expressions for the scalar thermostatic pressure and even-

tually other extra reversible contributions like reversible stress tensors. After

the evolution equations have been derived, we have no freedom to use other

physical requirements to cancel, add or modify the governing evolution re-

lations. The only thing that we can change is the Poisson bracket, indeed

we cannot exclude the possibility that other Poisson brackets, diﬀerent from

these used in the preceding sections, provide a better physical description of

the physical process.

How do we ﬁnd the Poisson brackets? A ﬁrst type of arguments consists

in proceeding by trials and errors and, guided by our physical intuition, by

276 10 Hamiltonian Formalisms

comparing the results as arising from the Poisson kinematics with results de-

rived from other approaches. A second more sophisticated method consists in

using the Lie group of transformations as addressed in

¨

Ottinger’s book (2005)

to which we refer for more details. The dissipative part of the evolution equa-

tions, as speciﬁed by bracket (10.20) is derived after an explicit expression for

the dissipation potential ψ has been determined, the latter being compelled

to obey the properties to be zero at equilibrium and to be a convex function

of the variables outside equilibrium. To establish a valuable expression for

the dissipation potential is, generally speaking, not an easy task as it should

involve material information related to non-equilibrium parameters as vis-

cosity, diﬀusion coeﬃcient, thermal conductivity, hydrodynamic interaction

tensors, etc.

In short, the formulation of the evolution equations of GENERIC requires

the knowledge of the following entities: the Poisson bracket, the energy E

(or an equivalent thermodynamic potential as Helmholtz free energy), the

dissipation potential ψ, and entropy S. The three functions E, S and ψ

contain the individual features of the medium under study while to be sure

that we are in presence of a true Poisson bracket, we have to check the

Jacobi’s identity ¦A, ¦B, C¦¦ + ¦B, ¦C, A¦ + ¦C, ¦A, B¦¦; this represents a

heavy and tedious task but is of importance because it expresses that the

reversible dynamics has a Hamiltonian structure. In recent publications about

GENERIC, there is a tendency in favour of the use of the following building

blocks: the two generators E (energy) and S (entropy) and the two operators

ˆ

L and

ˆ

M.

As shown in Sect. 10.3, the splitting of the total pressure tensor as

pδ

αβ

+ P

αβ

into a reversible part pδ

αβ

and an extra viscous tensor P

αβ

arises as a result of the GENERIC approach. This is not so in the con-

text of other formalisms as classical irreversible thermodynamics, extended

irreversible thermodynamics or rational thermodynamics, where this split-

ting is imposed a priori. Moreover, in most non-equilibrium thermodynamic

theories, it is the entropy production that is put into the focus; the part that

does not produce entropy, the so-called gyroscopic forces of the form ω x

where ω is the angular velocity, or the convected parts of the generalized

time derivatives remain largely undetermined. It is a success of GENERIC to

incorporate naturally such gyroscopic eﬀects thanks to the Poisson structure

of the evolution equations and to propose explicit relations for these non-

dissipative contributions. In that respect, it is worth noting that Jacobi’s

identity imposes that the convected derivatives fulﬁl the principle of material

frame-indiﬀerence; this exhibits the strong link between Jacobi’s identity and

frame-indiﬀerence and sheds a new lighting on this controversial principle (see

Sect. 9.5).

Note also that in GENERIC, even the well-known and well-tested balance

equations of mass, momentum, and energy are re-derived while such relations

are usually taken for granted in other thermodynamic theories. Moreover,

the governing equations in GENERIC are generally non-linear and this is the

10.3 Final Comments 277

reason why they ﬁnd a natural domain of application in rheology. Another

diﬀerence between GENERIC and some other thermodynamic approaches

is that in the latter, the Onsager reciprocal relations are only applied in

the linear regime; in GENERIC, this property is taken for granted even in

the non-linear domain as attested by the requirement that the symmetry of

operator

ˆ

M is to be universally satisﬁed.

It should also be realized that in GENERIC, it is generally understood

that the system under consideration is isolated from its environment as it is

assumed that the total energy is conserved dE/dt = 0 and the total entropy

is increasing with time dS/dt > 0. This means that the theory is not well

adapted for describing open and/or externally forced systems as for instance

B´enard’s convection in a ﬂuid submitted to an external temperature gradient.

In the case of non-isolated systems, one should, as in classical thermodynam-

ics, introduce some kind of hypothetical reservoirs but such an operation

may be delicate and obscure the formalism (Grmela 2001). In a recent paper,

¨

Ottinger (2006) suggests to split the Poisson and dissipative brackets into

bulk and boundary contributions.

Moreover, in presence of coupled heat and mass transport, the procedure

dictated by GENERIC is rather intricate and requires the introduction of

unusual state variables like some particular combinations of the entropy ﬂux

and the mass ﬂux (Grmela et al. 2003). Finally, similarly to other approaches,

GENERIC formalism remains silent about the physical meaning of some basic

quantities such as non-equilibrium temperature.

GENERIC has been the subject of several applications in statistical me-

chanics, molecular dynamics, quantum mechanics and macroscopic physics,

mainly in rheology (

¨

Ottinger 2005) where one of the most important prob-

lems is to derive constitutive equations consistent with the dynamics of the

basic variables. A number of new results produced by GENERIC should also

be mentioned, for instance the formulation of generalized reptation models

(

¨

Ottinger 1999a), of equations in relativistic hydrodynamics and cosmology

(Ilg and

¨

Ottinger 1999;

¨

Ottinger 1999b), new models for heat transfer in

nano-systems (Grmela et al. 2005). GENERIC may also be considered as

an extension of the bracket formulation as proposed by Beris and Edwards

(1994) and Beris (2003).

The essential merit of GENERIC is its unifying universality as it is applica-

ble whatever the level of description of the material system, either macro-

scopic, mesoscopic, or microscopic. It should also be realized that GENERIC,

as well as extended thermodynamics, internal variables theories or rational

thermodynamics, are not exclusive of each other: they complement harmo-

niously and oﬀer, in most circumstances, results that are in agreement with

each other. Finally, yet importantly, it is expected that a better comprehen-

sion of all these theories will provide a step forward the formulation of a

desirable uniﬁed theory of irreversible processes.

278 10 Hamiltonian Formalisms

10.4 Problems

10.1. Poisson bracket. Show that the scalar product (∂A/∂x,

ˆ

L∂B/∂x) with

ˆ

L =

_

0 1

−1 1

_

is equivalent to the Poisson bracket ¦A, B¦.

10.2. Jacobi’s identity. Verify that by taking

ˆ

L =

_

0 1

−1 0

_

, the bracket (10.4)

satisﬁes Jacobi’s identity (10.8).

10.3. Degeneracy conditions. Prove that the results dE/dt = 0, dS/dt >0

may be considered as a direct consequence of the degeneracy conditions

ˆ

LδS/δx = 0 and

ˆ

M δE/δx = 0.

10.4. Degeneracy conditions. Check that the operators

ˆ

L and

ˆ

M given by

(10.46) verify the degeneracy properties

ˆ

LδS/δx = 0,

ˆ

M δE/δx = 0.

10.5. Isothermal ﬂuid. Show that in the case of a one-component isothermal

ﬂuid characterized by the variables x = (ρ, u

α

), the functional derivatives

δE/δx and δS/δx are given by

δE

δx

=

_

−u

2

/2ρ +∂ε/∂ρ

u

α

/ρ

_

,

δS

δx

=

_

−µ/T

0

_

,

wherein µ designates the chemical potential and T the uniform temperature.

10.6. Non-isothermal ﬂuid. Derive the GENERIC building blocks E, S, L and

M for a one-component ﬂuid in presence of temperature gradients (

¨

Ottinger

and Grmela 1997).

10.7. Matter diﬀusion. Establish the relevant equations of matter diﬀusion

in binary mixtures in the simpliﬁed case of non-viscous ﬂuids and uniform

temperature.

10.8. Non-Fickian diﬀusion. Repeat the analysis of Sect. 10.2.2 when it is

assumed that the barycentric velocity u does not vanish.

10.9. Thermodiﬀusion. Consider the more complex problem of coupling of

heat and mass transport (thermodiﬀusion) in a two-component system. De-

termine the evolution equations for the total mass density, the mass fraction,

the momentum, and the energy in the framework of the above Hamiltonian

formalism. Hint: See Grmela et al. (2003).

10.10. Extended irreversible thermodynamics. Show that the evolution equa-

tions governing the behaviour of the dissipative ﬂuxes of extended irreversible

thermodynamics possess a Hamiltonian structure (Grmela and Lebon 1990).

Chapter 11

Mesoscopic Thermodynamic Descriptions

Thermodynamics and Fluctuations

The previous chapters were essentially concerned with the average behaviour

of thermodynamic variables without reference to their ﬂuctuations. The lat-

ter are not only a subject of intense study in statistical mechanics, but also

play a signiﬁcant role in non-equilibrium thermodynamics, like near critical

points, non-equilibrium instabilities and, more generally, in processes taking

part at small time and space scales. They played an important role in the

original derivation of Onsager–Casimir’s reciprocal relations, and in modern

formulations of the ﬂuctuation–dissipation relations expressing the transport

coeﬃcients in terms of the correlation functions of the ﬂuctuations of the

ﬂuxes. Furthermore, when the ﬂuctuations become important, they must be

included amongst the set of independent variables. The status of ﬂuctua-

tions being intermediate between the macroscopic and the microscopic, it is

justiﬁed to refer to them as mesoscopic variables.

In the ﬁrst part of this chapter is presented an introductory overview of the

theory of ﬂuctuations. After recalling Einstein’s theory of ﬂuctuations around

equilibrium, we derive Onsager–Casimir’s reciprocal relation, and discuss the

ﬂuctuation–dissipation theorem, which is important in non-equilibrium sta-

tistical mechanics. Afterwards, we analyse two formalisms which have in-

ﬂuenced the recent developments of irreversible thermodynamics, namely

Keizer’s theory, where a non-equilibrium entropy is deﬁned in terms of the

second moments of ﬂuctuations, and the so-called mesoscopic thermodynam-

ics, in which the distribution function of the ﬂuctuations is itself the central

variable.

11.1 Einstein’s Formula: Second Moments

of Equilibrium Fluctuations

In 1902, Einstein proposed a relation between the probability of the ﬂuctu-

ation of a thermodynamic variable with respect to its equilibrium value and

the entropy change in isolated systems. This was obtained by inverting the

279

280 11 Mesoscopic Thermodynamic Descriptions

well-known Boltzmann–Planck’s formula

S = k

B

ln W, (11.1)

where S is the total entropy of the system, W is the probability of the

macrostate, and k

B

is the Boltzmann constant. With Einstein, let us write

W ≈ exp(∆S/k

B

), (11.2)

where ∆S is the change of entropy associated to the ﬂuctuations, namely

∆S = S −S

eq

. (11.3)

As an example, consider the ﬂuctuations of energy in an isolated system

composed of two subsystems in thermal contact. In equilibrium, both of them

have the same temperature, and the total internal energy is distributed in

such a way that the total entropy is maximum. However, despite the total

internal energy remains constant, both subsystems may exhibit ﬂuctuations

in their respective internal energies.

If the ﬂuctuations are small, ∆S may be expanded around equilibrium, so

that

∆S ≈ (δS)

eq

+

1

2

(δ

2

S)

eq

, (11.4)

δS and δ

2

S being the ﬁrst and the second diﬀerentials of the entropy. For

an isolated system, one has (δS)

eq

= 0 and (δ

2

S)

eq

≤ 0, because the total

entropy is maximum at equilibrium. Introduction of these results into (11.2)

yields the well-known Einstein’s formula for the probability of ﬂuctuations,

W ≈ exp

_

1

2

δ

2

S

k

B

_

. (11.5)

Note that this relation is restricted to second-order developments and is only

applicable to isolated systems.

To extend (11.5) to non-isolated systems, Einstein suggested to express

the probability of ﬂuctuations as

W ≈ exp

_

−

∆/

k

B

T

0

_

, (11.6)

where /, the availability, is a new quantity deﬁned as

/ = U −T

0

S +p

0

V, (11.7)

in which T

0

and p

0

are the temperature and pressure of the environment.

Thus, / is not an intrinsic property of the system alone but of the ensemble

formed by it and its surroundings. It can be shown that ∆/ measures the

maximum amount of useful work which can be extracted from the system

during exchanges with the outside world. If the system and the surroundings

11.1 Einstein’s Formula: Second Moments of Equilibrium Fluctuations 281

are in thermal and mechanical equilibrium, T = T

0

and p = p

0

, the change

in / is given by

∆/ = ∆U −T∆S +p∆V = ∆G, (11.8)

where G is the Gibbs’ function, so that (11.6) takes the form

W ≈ exp(−∆G/k

B

T). (11.9)

Similarly, for a system at ﬁxed V and T, ∆/ = ∆F, with F the Helmholtz

free energy, and the probability of ﬂuctuations becomes

W ≈ exp(−∆F/k

B

T). (11.10)

For small ﬂuctuations, ∆/ can be expanded as

∆/ ≈ δ/

eq

+

1

2

(δ

2

/)

eq

+ ,

where the ﬁrst-order term is zero because at equilibrium G is minimum at

ﬁxed T and p (or, respectively, F at ﬁxed V and T). As a consequence,

W ≈ exp

_

−

1

2

δ

2

G

k

B

T

_

, (11.11a)

W ≈ exp

_

−

1

2

δ

2

F

k

B

T

_

. (11.11b)

These equations can still be expressed in terms of δ

2

S because δ

2

F at constant

T and V (or δ

2

G at ﬁxed p and T) is proportional to δ

2

S. Indeed, remember

that F = U−TS; when one studies the ﬂuctuations of F at constant T and V ,

U is the independent variable and S is a function of U so that δ

2

F = −Tδ

2

S.

From (11.11a), one recovers (11.5), but with δ

2

S computed at constant T

and V . Similarly, it is easy to see that δ

2

G = −Tδ

2

S when combined with

(11.11b), again leads to (11.5) but with δ

2

S computed at constant T and p.

These observations indicate that (11.5) is applicable even outside isolated

systems, at the condition to specify explicitly which variables are kept ﬁxed.

In an isolated system, information about the ﬂuctuations around equi-

librium states implies the knowledge of δ

2

S in terms of δU and δV . The

second variation of S is directly obtained from the Gibbs’ equation δS =

T

−1

δU +pT

−1

δV which, introduced in (11.5) yields

W ∝ exp

_

−

1

2k

B

_

(T

−1

)

U

(δU)

2

+ 2(T

−1

)

V

δUδV + (T

−1

p)

V

(δV )

2

¸

_

,

(11.12)

where subscripts U and V stand for partial diﬀerentiation with respect to

these variables.

Recall the mathematical result that for a Gaussian distribution function

of the form

W ≈ exp(−

1

2

E

ij

δx

i

δx

j

), (11.13)

282 11 Mesoscopic Thermodynamic Descriptions

the second moments of the ﬂuctuations are given by

¸δx

i

δx

j

) = (E

−1

)

ij

, (11.14)

where δx

i

designates the ﬂuctuation of x

i

and the brackets ¸. . .) denote the

average over the probability distribution (11.13). The second moments of the

ﬂuctuations of U and V may be obtained from (11.12) and (11.14) by taking

into account that the matrix corresponding to the second derivatives of the

entropy with respect to these variables is

E

UV

= −

1

k

B

_

(T

−1

)

U

(T

−1

)

V

(T

−1

p)

U

(T

−1

p)

V

_

. (11.15)

By inverting the matrix and introducing the result in (11.14), it is left as an

exercise to prove that (see Problem 11.1)

¸δUδU) = k

B

C

p

T

2

−2k

B

T

2

pV α +k

B

Tp

2

V κ

T

,

¸δUδV ) = k

B

T

2

V α −k

B

TpV κ

T

,

¸δV δV ) = k

B

TV κ

T

,

(11.16)

where C

p

is the heat capacity at constant pressure; α is the coeﬃcient of

thermal expansion, and κ

T

is the isothermal compressibility. Near critical

points, when some of the coeﬃcients in (11.16) diverge, the second moments

of the ﬂuctuations become very large and they have a clear macroscopic

inﬂuence on the behaviour of the system. The analysis of this behaviour has

been a main topic of research in statistical mechanics from the 1960s to 1980s

(e.g. Callen 1985; Landau and Lifshitz 1980; Reichl 1998).

There is a natural distinction in the ﬂuctuation theory between statics

and dynamics. The former is related to properties occurring at equal times,

as discussed in the present section, the latter is concerned with their evo-

lution during ﬁnite time intervals. This aspect is discussed in the next sec-

tions in relation with the derivation of Onsager–Casimir’s relations and the

dissipation–ﬂuctuation theorem.

11.2 Derivation of the Onsager–Casimir’s Reciprocal

Relations

The analysis of the evolution of ﬂuctuations in the course of time led On-

sager to establish his famous reciprocal relations, which were introduced as

a phenomenological postulate in Chap. 2, and whose microscopic derivation

is outlined here.

Assume that the entropy of an isolated system is a function of several

variables A

1

, . . . , A

n

and denote by α

β

the ﬂuctuations from their average

11.2 Derivation of the Onsager–Casimir’s Reciprocal Relations 283

equilibrium value ¸A

β

), i.e. α

β

= A

β

− ¸A

β

). For small ﬂuctuations around

equilibrium, the entropy may be written as

S = S

eq

−

1

2

G

βγ

α

β

α

γ

, (11.17)

with

G

βγ

=

_

∂

2

S

∂A

β

∂A

γ

_

eq

, (G

βγ

= G

γβ

). (11.18)

Summation with respect to repeated subindices is understood. It follows from

Einstein’s relation (11.5) that the second moments of the ﬂuctuations are

given by

¸α

β

α

γ

) = k

B

(G

−1

)

βγ

. (11.19)

In his original demonstration, Onsager assumed that the decay of spon-

taneous ﬂuctuations of the system obeys the same phenomenological laws

that describe the evolution of the perturbations produced by external causes.

Furthermore, he supposed that these laws are linear, so that the evolution of

α

β

is governed by

dα

β

dt

= −M

βγ

α

γ

(11.20)

with M

βγ

a matrix independent of α

γ

.

Moreover, Onsager deﬁnes the thermodynamic ﬂuxes J

β

as the time deriv-

atives of the ﬂuctuations α

β

and the thermodynamic forces X

β

as the deriv-

atives of the entropy with respect to α

β

, i.e.

J

β

=

dα

β

dt

, (11.21a)

X

β

=

_

∂∆S

∂α

β

_

= −G

βγ

α

γ

. (11.21b)

With these deﬁnitions of J

β

and X

β

, the time derivative of the entropy is

bilinear in the ﬂuxes and the forces

dS

dt

= −G

βγ

α

γ

dα

β

dt

= J

β

X

β

, (11.22)

while the evolution equations (11.20) can be cast in the form of linear relations

between ﬂuxes and forces:

J

β

= L

βγ

X

γ

. (11.23)

Indeed, it suﬃces to introduce the deﬁnitions (11.21) into (11.23) to obtain

dα

β

dt

= −L

βγ

G

γη

α

η

. (11.24)

Comparison with (11.20) leads to the identiﬁcation

L = M G

−1

. (11.25)

284 11 Mesoscopic Thermodynamic Descriptions

In a steady state, which by deﬁnition is invariant with respect to a translation

in time, one has for any couple of variables α

β

and α

γ

¸α

β

(0)α

γ

(t)) = ¸α

β

(0 +τ)α

γ

(t +τ)). (11.26)

Since this equality is valid for any values of t and τ; it is in particular true

for τ = −t, from which

¸α

β

(0)α

γ

(t)) = ¸α

β

(−t)α

γ

(0)). (11.27)

Assume that the variable A

β

has a well-deﬁned time-reversal parity under

the transformations t → −t, by deﬁnition, even variables will transform as

α

β

(−t) = α

β

(t), while odd variables will behave as α

β

(t) = −α

β

(−t). Mass

and energy are examples of even variables, and momentum and heat ﬂux are

odd variables. This allows us to write (11.27) in whole generality as

¸α

β

(0)α

γ

(t)) = ε

β

ε

γ

¸α

β

(t)α

γ

(0)), (11.28)

where ε

β

= +1 or −1 according the variable is even or odd. Note that (11.28)

implies that equal-time correlations between even and odd variables vanish

in equilibrium. Relation (11.28) expresses the microscopic reversibility of the

system and is of fundamental importance in deriving the Onsager–Casimir’s

relations.

Moreover, direct integration of the evolution equation (11.20) yields

α

β

(t) = exp[−Mt]

βγ

α

γ

(0). (11.29)

For small values of t, one may expand (11.29) in Taylor’s series, by keeping

only ﬁrst-order terms in t, and (11.28) takes then the form

¸α

β

(0)(δ

γη

−M

γη

t)α

η

(0)) = ε

β

ε

γ

¸(δ

βη

−M

βη

t)α

η

(0)α

γ

(0)), (11.30)

from which

M

γη

¸α

β

(0)α

η

(0)) = ε

β

ε

γ

M

βη

¸α

η

(0)α

γ

(0)). (11.31)

since δ

γη

= 1 if γ = η, δ

γη

= 0 otherwise. Expressing the second moments in

terms of the matrix G of the second derivatives of the entropy, (11.31) can

be written as

M

γη

(G

−1

)

βη

= ε

β

ε

γ

M

βη

(G

−1

)

ηγ

. (11.32)

From relation (11.25) between L and M, and the property that G is a sym-

metric matrix by construction, it follows that

L

γβ

= ε

β

ε

γ

L

βγ

, (11.33)

11.3 Fluctuation–Dissipation Theorem 285

which are the well-known Onsager–Casimir’s reciprocal relations. Historically,

Onsager derived the reciprocity relations in 1931, but only in the case of even

variables. The extension to odd variables was achieved in 1945 by Casimir.

The above analysis exhibits clearly the main assumptions underlying the

derivation of the Onsager–Casimir’s relations, namely linear evolution equa-

tions and the hypothesis that the behaviour of ﬂuctuations is described on the

average by relations mimicking the macroscopic transport laws. In Chap. 2

these relations were presented as a macroscopic postulate, whereas here it

follows from dynamical equations and statistical considerations.

In presence of a magnetic ﬁeld H, the operation of time reversal will only

change the internal magnetic ﬁeld arising from the motion of the particles,

which is automatically reversed under time reversal. However, because of

the Lorentz force, which is proportional to the vector product v H of the

velocities and the magnetic induction, the external magnetic ﬁeld H must

be inverted in order that Lorentz’s force remains unchanged; this guarantees

that the reversal motion follows the same path as the direct one. The same

comment applies when the system rotates as a whole with angular velocity

ω; the sign of the angular velocity must be changed because the Coriolis

force is proportional to the vector product ω v of the velocities of the

particles and the angular velocity of the system. Therefore, in presence of

an external magnetic induction H and angular rotation ω, the Onsager–

Casimir’s relations read as

L

γβ

(H, ω) = ε

β

ε

γ

L

βγ

(−H, −ω). (11.34)

The importance and the limitations of the Onsager–Casimir’s relations have

been underlined in Chap. 2. Here, the role of the time-reversal parity of the

several variables becomes clear, as well as the reason why H and ω must be

inverted to keep the validity of the reciprocal relations.

11.3 Fluctuation–Dissipation Theorem

In his paper on Brownian motion published in 1905, Einstein established that

molecular collisions between the ﬂuid and the Brownian particles are the

source of friction or dissipation and moreover responsible for the occurrence

of ﬂuctuations in the position of the particles. Einstein’s idea was further

explored by many authors, and was the basis for the derivation of the so-called

ﬂuctuation–dissipation theorem, expressing that ﬂuctuations and dissipative

coeﬃcients are related to each other. Langevin’s stochastic analysis of the

motion of a Brownian particle, and Nyquist study of the ﬂuctuations of the

electric current in a conductor contributed further to a better understanding

of the relation between ﬂuctuations and dissipation. Since the 1950s, these

analyses were set on a more general basis by Callen and Welton (1951),

286 11 Mesoscopic Thermodynamic Descriptions

Callen and Greene (1952), and Tisza and Manning (1957). It is not here our

purpose to enter into the details of their derivations, but rather to underline

the general ideas, because of their importance in statistical thermodynamics.

One of the main objectives of non-equilibrium statistical mechanics is to

provide, at the microscopic level, an expression for the memory function

relating a ﬂux J(t), at time t, to its conjugate force X(t

), at previous times

t

≤ t:

J(t) =

_

0

−∞

K(t −t

) X(t

)dt

; (11.35)

K(t −t

**) is the so-called memory function or memory kernel. As an example,
**

consider the Maxwell–Cattaneo evolution equation written formally as,

τ

∂J

∂t

= −(J −L X), (11.36)

with L the corresponding transport coeﬃcient. Integration of (11.36) yields,

for a zero initial ﬂux at time t

= −∞,

J(t) =

_

0

−∞

L

τ

exp [−(t −t

)/τ] X(t

)dt

. (11.37)

By comparing (11.37) and (11.35), one may identify the memory function as

K(t −t

) =

L

τ

exp

_

−

t −t

τ

_

, (11.38)

indicating that in the linearized version of extended irreversible thermo-

dynamics (EIT), the memory function decays exponentially. More general

versions of transport theory can be found in modern statistical mechanics,

based on N-particle distribution functions instead of the more restrictive

one-particle distribution function, and on the Liouville equation for the de-

scription of its evolution.

The most signiﬁcant feature of the ﬂuctuation–dissipation theorem is that

the memory kernel is related to the correlation function of the ﬂuctuations

of the corresponding ﬂuxes through (McQuarrie 1976; Reichl 1998)

K(t −t

) =

1

k

B

T

¸δJ(t)δJ(t

)), (11.39)

angular brackets stand for the average over an equilibrium distribution

function.

As particular examples of (11.39), the memory functions generalizing the

thermal conductivity and shear viscosity are written as

11.3 Fluctuation–Dissipation Theorem 287

λ

ij

(t −t

) =

V

k

B

T

2

¸δq

i

(t)δq

j

(t

)), (11.40)

η

ijkl

(t −t

) =

V

k

B

T

¸δP

v

ij

(t)δP

v

kl

(t

)). (11.41)

The knowledge of the quantities inside the brackets of (11.40) and (11.41)

in microscopic terms opens the way to the determination of the memory

functions in the most general situations, from dilute ideal gases to dense

ﬂuids.

In terms of the memory functions (11.40) and (11.41), the Fourier’s and

Newton–Stokes’ equations are generalized in the form

q

i

(t) = −

_

0

−∞

λ

ij

(t −t

)

∂T(t

)

∂x

j

dt

, (11.42a)

P

v

ij

(t) = −

_

0

−∞

η

ijkl

(t −t

)V

kl

(t

)dt

. (11.42b)

When the relaxation times of the ﬂuctuations of the ﬂuxes are very short, in

such a way that the memory functions are negligible except for very small

values of t −t

**, the bounds of the integrals may be extended to inﬁnity and
**

the thermodynamic forces may be taken out of the integral with the value

corresponding to t = t

**. Doing so, one recovers from (11.40) and (11.41), the
**

Green–Kubo’s formula relating the dissipative coeﬃcients of heat conductiv-

ity and shear viscosity to the ﬂuctuations of the heat ﬂux and the momentum

ﬂux, respectively:

λ

ij

= (k

B

T

2

)

−1

_

∞

0

¸δq

i

(0)δq

j

(t))dt, (11.43a)

η

ijkl

= (k

B

T)

−1

_

∞

0

¸δP

v

ij

(0)δP

v

kl

(t))dt. (11.43b)

The Green–Kubo’s relations are useful for calculating the dissipative trans-

port coeﬃcients from ﬁrst principles, independently of the density of the sys-

tem, in contrast with the classical calculations starting from the Boltzmann

equation, whose validity is limited to a low-density range (Reichl 1998).

Despite their elegance and generality, the simplicity of the above results

is misleading because of the complexity of the microscopic Liouville oper-

ator describing the evolution of the ﬂuctuations of the ﬂuxes. This is why

some approximations must be introduced, usually in the form of mathemat-

ical models for the memory functions, as a prerequisite for solving practical

problems. Therefore, formalisms such as EIT may be of valuable assistance

to model the evolution of the ﬂuctuations of the ﬂuxes.

288 11 Mesoscopic Thermodynamic Descriptions

11.4 Keizer’s Theory: Fluctuations in Non-Equilibrium

Steady States

11.4.1 Dynamics of Fluctuations

Keizer (1978, 1983, 1987) proposed to regard the ﬂuctuations as the founda-

tions for a deﬁnition of entropy outside equilibrium. More recently (Reguera

2000, Rubi 2004), attempts have been made to upgrade the ﬂuctuations to the

status of independent variables besides their usual average values. This topic

will be analysed in the next section.

Keizer combines the molecular picture of Boltzmann and the stochastic

picture of Onsager, and lays the foundations of a statistical thermodynam-

ics of equilibrium and non-equilibrium steady states. Steady states are, like

equilibrium states, time-independent but they involve constant inputs and

outputs per unit time, i.e. they require more parameters, like the ﬂuxes, to

specify the state.

Consider a large molecular system undergoing a series of elementary

processes characterized by a set of extensive variables x(= U(energy),

V, N . . . N (volume), N

1

, . . . , N

n

, with N

i

the number of particles of com-

ponent i). According to the Boltzmann–Planck’s postulate, the probability

of a ﬂuctuation δx around local equilibrium is given by

W(δx) ≈ exp(∆S/k

B

) = ¦(2π)

−(n+2)

det[(−S

eq

/k

B

)]¦

1/2

exp(δ

2

S

eq

/2k

B

),

(11.44)

with

(S

eq

)

ij

= (∂

2

S/∂x

i

∂x

j

)

eq

, δ

2

S

eq

= (∂

2

S/∂x

i

∂x

j

)

eq

δx

i

δx

j

, (11.45)

after performing a Taylor expansion around equilibrium and using the max-

imum entropy principle. Like in Sect. 11.1, δx stands for the ﬂuctuations of

the variables with respect to their equilibrium values (namely δx(r, t) ≡

x(r, t) −x

eq

≡ (δU, δV, δN

1

, . . . , δN

n

). From now on, we shall omit for sim-

plicity to write the time and position dependence of the variables.

A main feature of Keizer’s theory is that the elementary molecular

processes can be described by “canonical” kinetic evolution equations of the

form

dx

i

dt

=

k

ω

ki

(V

+

k

−V

−

k

) +J

i

(x, t) ≡ R

i

(x, t). (11.46)

The ﬁrst term in the right-hand side represents the dissipative part of the

evolution and J

i

the non-dissipative part, R

i

is a compact notation to express

the sum of both terms. The quantity ω

ki

stands for ω

ki

= x

+

ki

− x

−

ki

, with

11.4 Keizer’s Theory: Fluctuations in Non-Equilibrium Steady States 289

the + or − upper indices denoting the forward or backward variation of the

values of the extensive variable i in the elementary process k, ﬁnally V

±

k

is

given by

V

±

k

= Ω

±

k

exp

⎡

⎣

−

1

k

B

j

x

±

kj

_

∂S

∂x

j

_

⎤

⎦

, (11.47)

with Ω

±

k

the intrinsic forward and backward rates of the elementary k process.

It was shown by Keizer that the Fourier’s, Fick’s, and Newton–Stokes’ trans-

port laws as well as the chemical mass action law or the Boltzmann collision

operator may be written in the canonical form (11.46).

It follows from (11.46) that the dynamics of ﬂuctuations around a steady

state may be described by

dδx

i

dt

=

_

∂R

i

∂x

j

_

δx

j

+f

i

≡ H

ij

(x

ss

)δx

j

+f

i

, (11.48)

where superscript ss stands for steady state, and f

i

for a stochastic noise. By

deﬁnition, one has ¸f

i

) = 0 and

¸f

i

(t)f

j

(t

)) =

j

ω

ki

_

V

+

k

(x

ss

) +V

−

k

(x

ss

)

¸

ω

kj

δ(t −t

) ≡ γ

ij

(x

ss

)δ(t −t

).

(11.49)

It follows from these results that the probability of ﬂuctuations in a non-

equilibrium steady state may be written in the Gaussian form

W(δx) = [(2π)

−(n+2)

det σ

−1

] exp

_

−

1

2

δx

T

σ

−1

δx

_

, (11.50)

with σ the covariance matrix,

σ ≡ ¸δxδx)

ss

, (11.51)

given by the generalized ﬂuctuation–dissipation expression

H(x

ss

) σ + σ H

T

(x

ss

) = −γ(x

ss

). (11.52)

Since the matrices H and γ are known from (11.48) and (11.49), the non-

equilibrium covariance matrix (11.51) may directly be obtained from (11.52).

11.4.2 A Non-Equilibrium Entropy

An important pillar of Keizer’s theory is the deﬁnition of a non-equilibrium

entropy function S

K

by interpreting (11.50) in an analogous way to (11.44),

namely as

∂

2

S

K

∂x

i

∂x

j

≡ −k

B

(σ

−1

)

ij

. (11.53)

290 11 Mesoscopic Thermodynamic Descriptions

This is a diﬀerential equation for S

K

, which may be solved provided that

∂σ

ij

/∂x

k

= ∂σ

ik

/∂x

j

, a condition that was explicitly studied by Keizer. At

equilibrium, σ reduces to the equilibrium covariance matrix and S

K

becomes

identical to the equilibrium entropy. The physics underlying the above deﬁ-

nition is that the entropy S

K

is related to the ﬂuctuations in the extensive

variables in a way analogous to Boltzmann–Planck’s formula (11.1).

When expressed in terms of S

K

, the statistical distribution (11.53) is of

the form W ≈ exp(δ

2

S

K

/2k

B

) with

δ

2

S

K

=

ij

(∂

2

S

K

/∂x

i

∂x

j

)

ss

(x

i

−x

ss

i

)(x

i

−x

ss

i

). (11.54)

Since stability requirements impose that σ must be a positive deﬁnite matrix,

it follows from (11.53) that in stable steady states δ

2

S

K

≤ 0. Moreover, it was

shown by Keizer that the second variation of S

K

is a Lyapounov function,

from which

δ

2

S

K

≤ 0,

d

dt

δ

2

S

K

≥ 0. (11.55)

Since non-equilibrium states can only be sustained by the presence of ﬂuxes

(of energy, mass, etc.) between the system and reservoirs (of energy, mass,

etc.), it is assumed by Keizer that S

K

will depend not only on the intensive

variables x but also on the ﬂuxes f and, in addition, some intensive variables

Γ characterizing the reservoirs, so that

S

K

= S

K

(x; f , Γ). (11.56)

In analogy with classical thermodynamics, Keizer deﬁnes intensive vari-

ables conjugate to the extensive variables x

i

by

φ

i

=

∂S

K

∂x

i

. (11.57)

In particular, the intensive variables conjugate to the energy, volume and

number of moles are related to generalized temperature T, pressure p, and

chemical potentials µ

i

, in such a way that the generalized Gibbs’ equation

for S

K

will read as

dS

K

= T

−1

dU +pT

−1

dV −

i

µ

i

T

−1

dN

i

+

j

∂S

K

∂f

j

df

j

+

l

∂S

K

∂Γ

l

dΓ

l

.

(11.58)

Note that (11.58) contains a contribution in the ﬂuxes, which is analogous

to EIT – the latter lacking the last term, related to the properties of the

reservoirs.

Expanding S

K

(x; f , Γ) around equilibrium results in

S

K

(x; f , Γ) = S

eq

(x) −

i

f

i

v

i

(x; f , Γ), (11.59)

11.4 Keizer’s Theory: Fluctuations in Non-Equilibrium Steady States 291

where ν

i

is the “force” conjugated to f

i

. As a consequence, any intensive

variable, like for instance the temperature T conjugated to the internal energy

U through 1/T = ∂S

K

/∂U, will be given by

1

T

=

1

T

eq

−

i

f

i

∂v

i

∂U

. (11.60)

This result shares some features with EIT, as it exhibits the property that

the temperature is not equal to the (local) equilibrium temperature T

eq

, but

contains additional terms depending on the ﬂuxes (see Box 7.3).

An alternative expression of the generalized entropy, more explicit in the

ﬂuctuations and based on (11.53), is

S

K

(x; f , Γ) = S

eq

(x) −

1

2

k

B

ij

x

i

x

j

(σ

−1

−σ

−1

eq

)

ij

, (11.61)

where x

0

= U corresponds to the internal energy. It is directly inferred from

(11.61) that the temperature can be cast in the form

1

T

=

1

T

eq

−k

B

i

x

i

(σ

−1

i0

−σ

−1

i0,eq

). (11.62)

This result is important as it points out that non-equilibrium temperature

(the same reasoning remains valid for the pressure or the chemical potential)

is not identical to the local equilibrium one but is related to it through the

diﬀerence of the non-equilibrium and equilibrium correlation function of the

ﬂuctuations.

A spontaneous variation of the extensive variables x

i

will lead to a rate of

change of entropy given by

dS

K

dt

=

i

φ

i

dx

i

dt

. (11.63)

Using the property that for small deviations with respect to the steady state

φ

i

−φ

ss

i

=

j

_

∂

2

S

∂x

i

∂x

j

_

ss

(x

j

−x

ss

j

), (11.64)

(11.54) can be written as

δ

2

S

K

=

i

(φ

i

−φ

ss

i

)(x

i

−x

ss

i

) (11.65)

and, after diﬀerentiation with respect to time,

d

dt

δ

2

S

K

= 2

i

(φ

i

−φ

ss

i

)

dx

i

dt

≥ 0. (11.66)

292 11 Mesoscopic Thermodynamic Descriptions

Rearranging (11.66), one obtains

i

φ

i

dx

i

dt

≥

i

φ

ss

i

dx

i

dt

. (11.67)

The left-hand side of inequality (11.67) refers to the instantaneous values

of the intensive variables whereas the right-hand side involves their average

values in the steady state, but in virtue of (11.63), the left-hand side is also

the rate of change of entropy so that ﬁnally,

dS

K

dt

≥

i

φ

ss

i

dx

i

dt

, (11.68)

which was called by Keizer a generalized Clausius inequality because it gen-

eralizes Clausius inequality T

R

dS/dt > dQ/dt established for a system in

equilibrium with a reservoir at temperature T

R

.

Keizer’s theory has been the subject of numerous applications as ion

transport through biological membranes, isomerization reactions, ﬂuctuations

caused by electro-chemical reactions, light scattering under thermal gradients,

laser heated dimerization, etc. (Keizer 1987).

11.5 Mesoscopic Non-Equilibrium Thermodynamics

At short time and small length scales, the molecular nature of the systems

cannot be ignored. Classical irreversible thermodynamics(CIT) is no longer

satisfactory; indeed, molecular degrees of freedom that have not yet relaxed

to their equilibrium value will inﬂuence the global dynamics of the system,

and must be incorporated into the description, as done for example in ex-

tended thermodynamics or in internal variables theories. Unlike these ap-

proaches, which use as variables the average values of these quantities and

in contrast also with Keizer’s theory that adds, as additional variables, the

non-equilibrium part of the second moments of their ﬂuctuations, mesoscopic

non-equilibrium thermodynamics describes the system through a probability

distribution function P(x, t), where x represents the set of all relevant de-

grees of freedom remaining active at the time and space scales of interest

(Reguera 2004; Rub´ı 2004).

The main idea underlying mesoscopic non-equilibrium thermodynamics is

to use the methods of CIT to obtain the evolution equation for P(x, t). The

selected variables do not refer to the microscopic properties of the molecules,

as for instance in the kinetic theory, but are obtained from an averaging pro-

cedure, as in macroscopic formulations. It is in this sense that the theory

is mesoscopic, i.e. intermediate between macroscopic and microscopic de-

scriptions. It shares some characteristics with extended thermodynamics or

internal variables theories, like an enlarged choice of variables and statistical

theories like the statistical concept of distribution function.

11.5 Mesoscopic Non-Equilibrium Thermodynamics 293

11.5.1 Brownian Motion with Inertia

As illustration, we will consider the role of inertial eﬀects in the problem

of diﬀusion of N non-interacting Brownian particles of mass m immersed

in a ﬂuid of volume V . Inertial eﬀects are relevant when changes in spatial

density occur at timescales comparable to the time required by the velocity

distribution of particles to relax to its equilibrium Maxwellian value. At short

timescales, the particles do not have time to reach the equilibrium velocity

distribution, therefore the local equilibrium hypothesis cannot remain valid

and ﬂuctuations become relevant. They are modelled by introducing as extra

variables the probability density P(r, v, t) to ﬁnd the system with position

between r and r + dr and velocity between v and v + dv at time t. The

problem to be solved is to obtain the evolution equation of P(r, v, t).

The connection between entropy and probability of a state is given by the

Gibbs’ entropy postulate (e.g. de Groot and Mazur 1962), namely

s = s

eq

−

k

B

m

_

P(r, v, t) ln

P(r, v, t)

P

eq

(r, v)

dr dv, (11.69)

where s is the entropy per unit mass and P

eq

(r, v) the equilibrium distribu-

tion. By analogy with CIT, we formulate a Gibbs’ equation of the form

T ds = −

_

µ(r, v, t)dP(r, v, t)dr dv, (11.70)

where T is the temperature of a heat reservoir and µ the chemical potential

per unit mass, which can be given the general form

µ(r, v, t) = µ

eq

+

k

B

T

m

ln

P(r, v, t)

P

eq

(r, v)

+K(r, v), (11.71)

wherein K(r, v) is an extra potential which does not depend on P(r, v, t).

Since no confusion is possible between chemical potentials measured per unit

mass or per mole, we have omitted the horizontal bar surmounting “µ”. For

an ideal system of non-interacting particles in absence of external ﬁelds, sta-

tistical mechanics considerations suggest to identify K(r, v) with the kinetic

energy per unit mass

K(r, v) =

1

2

v

2

. (11.72)

The evolution equation for the distribution function will be given by the

continuity equation in the position and velocity space, namely

∂P

∂t

= −

∂J

r

∂r

−

∂J

v

∂v

, (11.73)

where J

r

and J

v

are the probability ﬂuxes in the r, v space. The diﬀusion

ﬂux of particles in the physical r coordinate space is directly obtained by

integration over the velocity space

294 11 Mesoscopic Thermodynamic Descriptions

¯

J

r

(r, t) =

_

vP(r, v, t)dv. (11.74)

The ﬂuxes J

r

and J

v

in (11.73) are not known a priori, and will be derived

by following the methodology of non-equilibrium thermodynamic methods,

i.e. by imposing the restrictions placed by the second law. By diﬀerentiating

(11.70) with respect to time and using (11.73), it is found that the rate of

entropy production can be written as

T

ds

dt

=

_ _

−J

r

∂µ

∂r

−J

v

∂µ

∂v

_

dr dv, (11.75)

after performing partial integrations and supposing that the ﬂuxes vanish

at the boundaries. As in classical irreversible thermodynamics, one assumes

linear relations between the ﬂuxes J and the forces, so that

J

r

= −L

rr

∂µ

∂r

−L

rv

∂µ

∂v

, (11.76)

J

v

= −L

vr

∂µ

∂r

−L

vv

∂µ

∂v

, (11.77)

where L

ij

are phenomenological coeﬃcients, to be interpreted later on. To

ensure the positiveness of the entropy production (11.75), the matrix of these

coeﬃcients must be positive deﬁnite. Furthermore, if we take for granted the

Onsager–Casimir’s reciprocity relations, one has L

rv

= −L

vr

, with the minus

sign because r and v have opposite time-reversal parity.

To identify the phenomenological coeﬃcients, substitute (11.71) in (11.76)

and impose the condition that the particle diﬀusion ﬂux in the r-space should

be recovered from the ﬂux in the r, v space, namely

¯

J

r

(r, t) =

_

vP(r, v, t)dv =

_

J

r

(r, v, t)dv

= −

_ _

L

rr

k

B

T

m

1

P

∂P

∂r

+L

rv

k

B

T

m

1

P

∂P

∂v

+L

rv

v

_

dv. (11.78)

Since P(r, v, t) is arbitrary, (11.78) may only be identically satisﬁed if L

rr

=0,

L

rv

= −P, and the only left undetermined coeﬃcient is L

vv

. If it is taken as

L

vv

= P/τ, where τ is a velocity relaxation time related to the inertia of the

particles, (11.76) and (11.77) become, respectively,

J

r

=

_

v +

D

τ

∂

∂v

_

P, (11.79)

J

v

= −

_

D

τ

∂

∂r

+

v

τ

+

D

τ

2

∂

∂v

_

P, (11.80)

where D ≡ (k

B

T/m)τ is identiﬁed as the diﬀusion coeﬃcient. When these

expressions are introduced into the continuity equation (11.73), it is found

that

11.5 Mesoscopic Non-Equilibrium Thermodynamics 295

∂P

∂t

= −

∂

∂r

(vP) +

∂

∂v

_

v

τ

+

D

τ

2

∂

∂v

_

P. (11.81)

This is the well-known Fokker–Planck’s equation for non-interacting Brownian

particles in presence of inertia and it is worth to stress that this equation has

been derived only by thermodynamic methods, without explicit reference to

statistical mechanics. Equilibrium situations corresponds to the vanishing of

the ﬂuxes, i.e. J

r

= 0, J

v

= 0 and a Gaussian probability distribution. In

CIT, it is assumed that the probability distribution is the same as in equi-

librium, i.e. Gaussian, centred at a non-zero average, and with its variance

related to temperature in the same way as in equilibrium. Moreover, diﬀu-

sion in the coordinate r-space is much slower than in the velocity v-space,

so that it is justiﬁed to put J

v

= 0. From (11.80), it is then seen that

(v + (D/τ)∂/∂v)P = −D∂P/∂r which, substituted in (11.79), yields Fick’s

law

J

r

= −D

∂µ

∂r

. (11.82)

Far from equilibrium, neither the space nor the velocity distributions corre-

spond to equilibrium and one has J

r

,= 0, J

v

,= 0. In this case, the velocity

distribution may be very diﬀerent from a Gaussian form, and it is not clear

how to deﬁne temperature (see, however, Box 11.1 where an attempt to deﬁne

a non-equilibrium temperature is presented).

Box 11.1 Non-Equilibrium Temperature

In the present formalism, it is not evident how to deﬁne a temperature

outside equilibrium. To circumvent the problem, a so-called eﬀective tem-

perature has been introduced. It is deﬁned as the temperature at which

the system is in equilibrium, i.e. the one corresponding to the probability

distribution at which the rate of entropy production vanishes. Substituting

(11.79) and (11.80) in (11.75), it is easily proven that the entropy produc-

tion can be written as

ds

dt

=

_

P

Tτ

_

v +k

B

T

∂

∂v

ln P

_

2

drdv, (11.1.1)

from that the eﬀective temperature will be given by

1

T

eﬀ

= −

k

B

v

∂

∂v

ln P. (11.1.2)

This expression can be rewritten in a form recalling that of the equilibrium

temperature, i.e.

1

T

eﬀ

=

∂s

eﬀ

∂e

, (11.1.3)

at the condition to deﬁne an “eﬀective” entropy by s

eﬀ

= −k

B

ln P and an

energy density by e = (1/2)v

2

. Other deﬁnitions of eﬀective temperature are

296 11 Mesoscopic Thermodynamic Descriptions

possible. Indeed, by taking e = (1/2)(v − ¸v))

2

, where ¸v

2

) is the average

value of v

2

, the temperature would be the local equilibrium one, but this

would not correspond to a zero entropy production. Moreover, as mentioned

by Vilar and Rub´ı (2001), the eﬀective temperature is generally a function of

r, v, t; this means that, at a given position in space, there is no temperature

at which the system can be at equilibrium, because T(r, v, t) ,= T(r, t). If

it is wished to deﬁne a temperature at a position x, it would depend on the

way the additional degrees of freedom are eliminated.

One more remark is in form. It is rather natural to expect that the

mesoscopic theory discussed so far will cope with evolution equations of the

Maxwell–Cattaneo type, as the latter involve characteristic times compara-

ble to the relaxation time for the decay of the velocity distributions towards

its equilibrium value. Indeed by multiplying the Fokker–Planck’s equation

(11.81) by v and integrating over v, one obtains

τ

∂

¯

J

∂t

= −

¯

J −D∇n, (11.83)

where

¯

J is the diﬀusion ﬂux in space, given by (11.64), n(r, t) =

_

P(r, v, t)dv

the particle number density, and D = ¸v

2

)τ = (k

B

T/m) τ. This indicates

that mesoscopic non-equilibrium thermodynamics is well suited to describe

processes governed by relaxation equations of the Maxwell–Cattaneo type,

just like EIT. But now, it is the probability distribution function which is

elevated to the status of variable as basic variable rather than the average

value of the diﬀusion ﬂux. It should be added that working in the frame of

EIT allows also obtaining the second moments of the ﬂuctuations of J by

combining the expression of extended entropy with Einstein’s relation (11.5).

The above considerations could leave to picture that mesoscopic thermody-

namics is more general than EIT, however, when the second moments of

ﬂuctuations are taken as variables, besides their average value, EIT provides

an interesting alternative more easier to deal with in practical situations.

11.5.2 Other Applications

The above results are directly generalized when more degrees of freedom

than r and v are present. The probability density will then be a function

of the whole set of degrees of freedom, denoted x, so that P = P(x, t). The

analysis performed so far can be repeated by replacing in all the mathematical

expressions the couple r, v by x. In particular, the continuity equation will

take the form

∂P

∂t

= −

∂J

∂x

, (11.84)

11.5 Mesoscopic Non-Equilibrium Thermodynamics 297

where J is given by the constitutive relation

J = −

L

T

∂µ

∂x

. (11.85)

After substitution of (11.85) in (11.84) and use of (11.71) for the chemical

potential, one obtains the kinetic equation

∂P

∂t

=

∂

∂x

_

D

∂P

∂x

+

D

k

B

T

∂Φ

∂x

P

_

, (11.86)

which is a generalization of the Fokker–Planck’s equation (11.81), where now

Φ is not the kinetic energy as in (11.72) but it includes the potential energy

related to the internal degrees of freedom.

Extension to non-linear situations, as in chemical reactions, does not raise

much diﬃculty. If chemical processes are occurring at short timescales, they

will generally take place from an initial to a ﬁnal state through intermediate

molecular conﬁgurations. Let the variable x characterize these intermediate

states. The chemical potential is still given by (11.71) in which K is, for

instance, a bistable potential whose wells correspond to the initial and ﬁ-

nal states while the maximum represents the intermediate barrier. Such a

description is applicable to several problems as active processes, transport

through membranes, thermionic emission, adsorption, nucleation processes

(Reguera et al. 2005). Let us show, in particular, that mesoscopic thermody-

namics leads to a kinetic equation where the reaction rate satisﬁes the mass

action law. The linear constitutive law (11.85) is generalized in the form

J = −k

B

L

1

z

∂z

∂x

, (11.87)

where z ≡ exp(µ/k

B

T) is the fugacity; an equivalent expression is

J = −D

∂z

∂x

, (11.88)

where D ≡ k

B

L/z represents the diﬀusion coeﬃcient. Assuming D constant

and integrating (11.88) from the initial state 1 to the ﬁnal state 2 yields

¯

J ≡

_

2

1

J dx = −D(z

2

−z

1

) = −D

_

exp

µ

2

k

B

T

−exp

µ

1

k

B

T

_

, (11.89)

where

¯

J is the integrated rate. Expression (11.89) can alternatively be cast

in the more familiar form of a kinetic law

¯

J = K[1 −exp(−//k

B

T)], (11.90)

where K stands for Dexp(µ

1

/k

B

T) and / = µ

2

− µ

1

is the aﬃnity of the

reaction. When µ

i

/k

B

T ¸ 1, one recovers from (11.90), the classical linear

phenomenological law of CIT (see Chap. 4)

298 11 Mesoscopic Thermodynamic Descriptions

¯

J =

D

k

B

T

/. (11.91)

Up to now, the theory has been applied to homogeneous systems char-

acterized by the absence of gradients of thermal and mechanical quantities.

The constraint of uniform temperature is now relaxed and the question is

how to incorporate thermal gradients in the formalism. For the problem of

Brownian motion, this is achieved by writing the Gibbs’ equation as follows:

ds =

1

T

du −

1

T

_

µdP dx dv, (11.92)

where u is the energy density. The corresponding constitutive equations are

now given by (Reguera et al. 2005)

J

q

= −L

TT

∇T/T

2

−

_

k

B

L

Tv

∂

∂v

ln

P

P

leq

dv, (11.93)

J

v

= −L

vT

∇T/T

2

−

_

k

B

L

vv

∂

∂v

ln

P

P

leq

dv, (11.94)

where J

q

is the heat ﬂux, J

v

the probability current, and P

leq

the local equi-

librium distribution function; L

ij

are phenomenological coeﬃcients forming

a positive deﬁnite matrix and obeying the Onsager relation L

Tv

= −L

vT

.

Equations (11.93) and (11.94) exhibit clearly the coupling between the two

irreversible processes occurring in the system: diﬀusion probability and heat

conduction. The corresponding evolution equation of the probability density

is now

∂P

∂t

= −v ∇P +β

∂

∂v

_

Pv +k

B

T

∂P

∂v

_

+

γ

T

∂

∂v

(P∇T), (11.95)

where β is the friction coeﬃcient of the particles and γ a coeﬃcient related

to L

vT

.

To summarize, mesoscopic non-equilibrium thermodynamics has shown

its applicability in a wide variety of situations where local equilibrium is

never reached, as for instance relaxation of polymers or glasses, dynamics of

colloids or ﬂows of granular media. Such systems are characterized by internal

variables exhibiting short length scales and slow relaxation times. The basic

idea of this new theory is to incorporate in the description those variables,

which have not yet relaxed to their local equilibrium values. As illustrative

examples, we have discussed the problem of inertial eﬀects in diﬀusion of

Brownian particles where change of density take place at a timescale of the

order of the time needed by the particles to relax to equilibrium, chemical

reactions and thermodiﬀusion. The tools are borrowed from CIT (see Chap. 2)

but the original contribution is to introduce, amongst the set of variables, the

density probability distribution whose time evolution is shown to be governed

by a Fokker–Planck’s equation.

11.6 Problems 299

11.6 Problems

11.1. Second moments. (a) Show, from (11.12) and (11.15) that the second

moments of the energy and the volume ﬂuctuations are given by

¸δUδU) = −(k

B

/M)(∂U/∂T

−1

)

T

−1

p

,

¸δUδV ) = −(k

B

/M)(∂V/∂T

−1

)

T

−1

p

,

¸δV δV ) = −(k

B

/M)(∂V/∂T

−1

p)

T

−1.

(b) Write explicitly the second-order derivatives appearing in (11.12) and

(11.15), and derive expressions (11.16).

11.2. Density ﬂuctuations. (a) From the second moments of the volume ﬂuc-

tuations, write the expression for the density ﬂuctuations of a one-component

ideal gas at 0

◦

C and 1 atm, when the root-mean-square deviation in density

is 1% of the average density of the system? (b) Show that near a critical

point, where (∂p/∂V )

T

= 0, these ﬂuctuations diverge.

11.3. Dielectric constant. The dielectric constant ε of a ﬂuid varies with the

mass density according to the Clausius–Mossoti’s relation

ε −1

ε + 2

= Cρ,

with C is a constant related to the polarizability of the molecules and ρ is

the mass density. Show that the second moments of the ﬂuctuations of ε are

given by

¸(δε)

2

) =

k

B

Tκ

T

9V

(ε −1)

2

(ε + 2)

2

,

with κ

T

being the isothermal compressibility.

11.4. Density ﬂuctuations and non-locality. The correlations in density ﬂuc-

tuations at diﬀerent positions are usually written as ¸δn(r

1

)δn(r

2

)) ≡

¯ nδ(r

1

− r

2

) + ¯ nν(r) with ¯ n the average value of the density, r ≡ [r

2

−r

1

[,

and ν(r) the correlation function. To describe such correlations, Ginzburg

and Landau propose to include in the free energy a non-local term of the

form

F(T, n, ∇n) = F

eq

(T, n) −

1

2

b(∇n)

2

=

1

2

a(n − ¯ n)

2

−

1

2

b(∇n)

2

,

with a(T) a function of temperature which vanishes at the critical point and

b > 0 a positive constant. (a) If the density of ﬂuctuations are expressed as

n − ¯ n =

k

n

k

e

ik·r

,

300 11 Mesoscopic Thermodynamic Descriptions

show that

¸[δn

k

[

2

) =

k

B

T

V (a +bk

2

)

.

(b) Taking into account that

_

ν(r)e

−ik·r

dV =

V

n

¸[δn

k

[

2

) −1,

prove that

ν(r) =

k

B

Ta

4π¯ nb

1

r

exp

_

−

r

ξ

_

,

where ξ is a correlation length, given by ξ = (b/a)

1/2

. (Note that in the limit

ξ →0, one obtains ν(r) = δ(r), with δ(r) the Dirac’s function, and that near

a critical point the correlation length diverges.)

11.5. Transport coeﬃcients. (a) Apply (11.43) to obtain the classical results

η = nk

B

Tτ and λ = (5k

B

T

2

/2m)τ. In terms of the peculiar molecular ve-

locities c the microscopic operators for the ﬂuxes are given by

ˆ

P

v

12

= mc

1

c

2

and ˆ q

1

=

_

1

2

mc

2

−

5

2

k

B

T

_

c

1

. The equilibrium average should be performed

over the Maxwell–Boltzmann distribution function. (b) Make a similar analy-

sis for the memory kernel corresponding to the diﬀusion coeﬃcient, namely

D(t −t

) = (V/k

B

T) ¸δJ

i

(t)δJ

i

(t

)).

11.6. Second moments and EIT. Apply Einstein’s relation (11.5) to the en-

tropy (7.61) of EIT, and ﬁnd the second moments of the ﬂuctuations of the

ﬂuxes around equilibrium. Note that the results coincide with (11.43), ob-

tained from Green–Kubo’s relations by assuming an exponential relaxation

of the ﬂuctuations of the ﬂuxes.

11.7. Fluctuations around steady states. Assume with Keizer that Einstein’s

relation remains valid around a non-equilibrium steady state. Determine the

second moments of the ﬂuctuations of u and q around a non-equilibrium

steady state characterized by a non-vanishing average heat ﬂux q

0

. Compare

the results with those obtained in equilibrium in Problem 11.6.

11.8. Brownian motion. Langevin proposed to model the Brownian motion

of the particles by adding to the hydrodynamic friction force −ζv (ζ is the

friction coeﬃcient, v is the speed of the particle), a stochastic force f de-

scribing the erratic forces due to the collisions of the microscopic particles of

the solvent, in such a way that

m

dv

dt

= −ζv +f .

He assumed that f is white (without memory) and Gaussian, in such a way

that ¸f ) = 0, ¸f (t)f (t +t

)) = Bδ(t

**). Using the result that in the long-time
**

limit the equipartition condition ¸

1

2

mv

2

) =

3

2

k

B

T must be satisﬁed, show

11.6 Problems 301

that ¸f (t)f (t +t

)) = (ζ/m)k

B

Tδ(t

), i.e. B = (ζ/m)k

B

T. This is similar to a

ﬂuctuation–dissipation relation, as it shows that the second moments of the

ﬂuctuating force are proportional to the friction coeﬃcient. This is called a

ﬂuctuation–dissipation relation of the second kind.

11.9. Generalized ﬂuctuation–dissipation relation. The result of the previous

problem may be written in a more general way for any set a(t) of random

variables satisfying a linear equation of the form da/dt + H a = f with

H a friction matrix and f a white noise such that ¸f ) = 0, ¸f (t)f

T

(t +

t

)) = Bδ(t

**). Show that the matrix Bmust satisfy the ﬂuctuation–dissipation
**

relation

H G+G H

T

= B,

with G ≡ ¸δa δa

T

) the matrix of the second moments of ﬂuctuations of a(t).

11.10. Non-equilibrium temperature. To underline the connection between

the non-equilibrium temperature in Keizer’s formalism and in EIT, note that,

when the heat ﬂux q is the only relevant ﬂux, EIT predicts that ¸δuδu) =

¸δuδu)

eq

+α

q

2

, where subscript eq mean local equilibrium. In this expression

δu is the ﬂuctuations of the internal energy with respect to its steady state

average and α

**a coeﬃcient, whose explicit form is given in Jou et al. (2001).
**

The above result can still be cast in the form

q

2

= [¸δuδu) −¸δuδu)

eq

] (α

)

−1

.

Recalling that in EIT the non-equilibrium temperature T is given by

1

T

=

1

T

eq

−

1

2

∂α

∂u

q

2

,

write this expression in terms of the second moments of energy ﬂuctuations

and compare with Keizer’s expression (11.60).

Epilogue

By writing this book, our objective was threefold:

1. First, to go beyond equilibrium thermodynamics. Although it is widely

recognized that equilibrium thermodynamics is a universal and well-

founded discipline with many applications mainly in chemistry and en-

gineering, it should be realized that its domain of application is limited

to equilibrium states and idealized reversible processes, excluding dissi-

pation. This is suﬃcient to predict the ﬁnal equilibrium state, knowing

the initial state, but it is silent about the duration and the nature of the

actual process between the initial and the ﬁnal equilibrium states, whence

the need to go beyond equilibrium thermodynamics. Another reason is

that, to foster the contact between micro- and macroscopic approaches, it

is imperative to go beyond equilibrium as most of the microscopic theories

deal with situations far from equilibrium. But the problem we are faced

with is that the avenue of equilibrium thermodynamics bifurcates in many

routes.

2. Our second objective was to propose a survey, as complete as possible,

of the many faces of non-equilibrium thermodynamics. For pedagogical

reasons, we have restricted the analysis to the simplest situations, empha-

sizing physical rather than mathematical aspects. The presentation of each

theory is closed by a critical discussion from which can be concluded that

none of the various approaches is fully satisfactory. It appears that each

school has its own virtues but that, in practical situations, one of them

may be preferable to another. It is our purpose that after gone through the

present book, the reader will be able to make up his personal opinion. We

do not pretend to have been everywhere fully objective and exhaustive.

We have deliberately been silent about some valuable descriptions, as the

entropy-free theory of Meixner (1973a, b), the Lagrangian formalism of

Biot (1970), the variational analysis of Sieniutycz (1994), the thermody-

namics of chaos (Beck and Schl¨ogl 1993; Berdichevsky 1997; Gaspard 1998;

Ruelle 1991), the statistical approach of Luzzi et al. (2001, 2002), and

303

304 Epilogue

Tsallis’ non-extensive entropy formalism (2004). For reasons of place and

unity, and despite their intrinsic interest, we have also deliberately omitted

microscopic formulations, such as statistical mechanics (e.g. Grandy 1987),

kinetic theory (e.g. Chapman and Cowling 1970), information theory

(Jaynes 1963), molecular dynamics (Evans and Morriss 1990), and other

kinds of computer simulations (e.g. Hoover 1999; Hutter and J¨ ohnk 2004).

We apologize for these omissions; the main reason was our option to conﬁne

the volume of the book to a reasonable size rather than to write an exten-

sive encyclopaedia. We have also bypassed some approaches either because

of their more limited impact in the scientiﬁc community, or because they

lack of suﬃciently new fundamental ideas or techniques. Concerning this

multiplicity, it may be asked why so many thermodynamics? A tentative

answer may be found in the diversity of thought of individuals, depending

on their roots, environment, and prior formation as physicists, mathe-

maticians, chemists, engineers, or biologists. The various thermodynamic

theories are based on diﬀerent foundations: macroscopic equilibrium ther-

modynamics, kinetic theory, statistical mechanics, or information theory.

Other causes of diversity may be found in the selection of the most

relevant variables and the diﬃculty to propose an undisputed deﬁni-

tion of temperature, entropy, and the second law outside equilibrium.

At the exception of classical irreversible thermodynamics (CIT), it is

generally admitted that non-equilibrium entropy depends, besides clas-

sical quantities as mass, energy, charge density, etc., on extra variables

taking the form of dissipative ﬂuxes in extended irreversible thermody-

namics (EIT), internal structural variables in internal variables theories

(IVT) and in GENERIC and probality distribution function in meso-

scopic theories. Out of equilibrium, the constitutive relations can either

be cast in the form of linear algebraic phenomenological relations as in

CIT, integrals involving the memory as in rational thermodynamics (RT),

or time evolution diﬀerential equations as in EIT, IVT, and GENERIC.

The next natural question is then: what is the best approach? Although the

present authors have their own (subjective) opinion, we believe that the

ﬁnal answer should be left to each individual reader but there is no doubt

that trying to reach unanimity remains a tremendous challenging task.

3. We failed to meet a third objective, namely to bring a complete unity

into non-equilibrium thermodynamics. Being aware about the role of non-

uniformity and the importance of diversity, we realize that the achievement

of such a unity may appear as illusory. However, we do not think that it

is a completely desperate task; indeed, it is more than a dream to believe

that in a near future it would be possible to summon up all the pieces

of the puzzle and to build up a well-shaped, unique, and universal non-

equilibrium thermodynamics. In that respect, we would like to stress that

there exists a wide overlapping between the diﬀerent schools. More specif-

ically, all the theories contain as a special case the classical irreversible

thermodynamics. The Cattaneo model of heat conduction is not typical of

Epilogue 305

EIT but can also be obtained in the framework of IVT, RT, GENERIC,

and the mesoscopic description. It is our hope that the present book will

contribute to promote reconciliation among the several approaches and

foster further developments towards deeper and more uniﬁed formulations

of thermodynamics beyond equilibrium.

By the way, it was also our purpose to convince the reader that “thermo-

dynamics is the science of everything”. Clearly, thermodynamics represents

more than converting heat into work or calculating engine eﬃciencies. It is

a multi-disciplinary science covering a wide variety of ﬁelds ranging from

thermal engineering, ﬂuid and solid mechanics, rheology, material science,

chemistry, biology, electromagnetism, cosmology to economical, and even so-

cial sciences.

Among the several open and challenging problems, let us mention three of

them. The ﬁrst one is related to the limits of applicability of thermodynamics

to small systems, like found in nano-technology and molecular biophysics. At

the opposite, in presence of long-range interactions, such as gravitation, it

may be asked how the second law should be formulated when these eﬀects

are dominant, like in cosmology. Finally, does thermodynamics conﬂict with

quantum mechanics? how to reconcile the reversible laws of quantum the-

ory and the subtleties of quantum entanglement of distant systems with the

irreversible nature of thermodynamics?

By writing this book, we were guided by the intellectual ambition to better

understand the frontiers and perspectives of the multi-faced and continuously

changing domain of our knowledge in non-equilibrium thermodynamics. This

remains clearly an unﬁnished task and we would like to think to have con-

vinced the reader that, therefore, this fascinating story is far from

The End

References

Agrawal D.C. and Menon V.J., Performance of a Carnot refrigerator at maximum

cooling power, J. Phys. A: Math. Gen. 23 (1990) 5319–5326

Alvarez X. and Jou D., Memory and nonlocal eﬀects in heat transport: from diﬀusive

to ballistic regimes, Appl. Phys. Lett. 90 (2007) 083109 (3 pp)

Andresen B., Finite-time thermodynamics in simulated annealing, in Entropy and

Entropy Generation (Shiner J.S., ed.), Kluwer, Dordrecht, 1996

Angulo-Brown F., An ecological optimization criterion for ﬁnite-time heat engines,

J. Appl. Phys. 69 (1991) 7465–7469

Anile A.M. and Muscato O., An improved thermodynamic model for carrier transport

in semiconductors, Phys. Rev. B 51 (1995) 16728–16740

Anile A.M. and Pennisi S., Extended thermodynamics of the Blotekjaer hydrodynam-

ical model for semiconductors, Continuum Mech. Thermodyn. 4 (1992) 187–197

Anile A.M., Allegretto W. and Ringhofer C., Mathematical Problems in Semi-

Conductor Physics, Lecture Notes in Mathematics, vol. 1823, Springer, Berlin

Heidelberg New York, 2003

Astumian R.D., Thermodynamics and kinetics of a Brownian motor, Science 276

(1997) 917–922

Astumian R.D. and Bier M., Fluctuation driven ratchets. Molecular motors, Phys.

Rev. Lett. 72 (1994) 1766–1769

Bampi F. and Morro A., Nonequilibrium thermodynamics: a hidden variable

approach, in Recent Developments in Non-Equilibrium Thermodynamics (Casas-

V´ azquez J., Jou D. and Lebon G., eds.), Lecture Notes in Physics, vol. 199,

Springer, Berlin Heidelberg New York, 1984

Beck C. and Schl¨ogl F., Thermodynamics of Chaotic Systems, Cambridge University

Press, Cambridge, 1993

Bedeaux D., Mazur P. and Pasmanter R.A., The ballast resistor; an electro-thermal

instability in a conducting wire I; the nature of the stationary states, Physica A

86 (1977) 355–382

Bejan A., Entropy Generation Minimization, CRC, Boca Raton, FL, 1996

B´enard H., Les tourbillons cellulaires dans une nappe liquide transportant la chaleur

par conduction en r´egime permanent, Rev. G´en. Sci. Pures et Appliqu´ees 11 (1900)

1261–1271, 1309–1318

Berdichevsky V.L., Thermodynamics of Chaos and Order, Longman, Essex, 1997

Berg´e P., Pomeau Y. and Vidal C., L’ordre dans le chaos, Hermann, Paris, 1984

Beris A.N., Simple Non-equilibrium thermodynamics application to polymer rheology,

in Rheology Reviews (Binding D.N. and Walters K., eds.), pp 33–75, British Society

of Rheology, Aberystwyth, 2003

307

308 References

Beris A.N. and Edwards B.J., Thermodynamics of Flowing Systems with Internal

Microstructure, Oxford Science Pubisher, Oxford, 1994

Biot M.A., Variational Principles in Heat Transfer, Oxford University Press, Oxford,

1970

Bird R.B. and de Gennes P.G., Discussion about the principle of objectivity, Physica

A 118 (1983) 43–47

Bird R.B., Armstrong R.C. and Hassager D., Dynamics of Polymeric Liquids, 2nd

ed., vol. 1: Fluid Mechanics, Wiley, New York, 1987a

Bird R.B., Curtiss C.F., Armstrong R.C. and Hassager D., Dynamics of Polymeric

Liquids, 2nd ed., vol. 2: Kinetic Theory, Wiley, New York, 1987b

Block M.J., Surface tension as the cause of B´enard’s cells and surface deformation in

a liquid ﬁlm, Nature 178 (1956) 650

Bodenshatz E., Pech W. and Ahlers G., Recent developments in Rayleigh-B´enard

convection, Annu. Rev. Fluid Mech. 32 (2000) 851–919

Bragard J. and Lebon G., Nonlinear Marangoni convection in a layer of ﬁnite depth,

Europhys. Lett. 21 (1993) 831–836

Braun D. and Libchaber A., Phys. Rev. Lett. 89 (2002) 188103

Brenner H., Rheology of two-phase systems, Annu. Rev. Fluid Mech. 2 (1970) 137–176

Bricmont J., Science of chaos or chaos in science? Physicalia Mag. 17 (1995) 159–221

Bridgman P.W., The Nature of Thermodynamics, Harvard University Press,

Cambridge, 1941

Busse F.H., Non-linear properties of thermal convection, Rep. Prog. Phys. 41 (1978)

1929–1967

Bustamante C., Liphardt J. and Ritort F., The nonequilibrium thermodynamics of

small systems, Phys. Today 58 (2005) 43–48

Callen H.B., Thermodynamics and an Introduction to Thermostatistics, 2nd ed.,

Wiley, New York, 1985

Callen H.B. and Welton T.A., Irreversibility and generalized noise, Phys. Rev. 83

(1951) 34–40

Callen H.B. and Greene R.F., On a theorem of irreversible thermodynamics, Phys.

Rev. 86 (1952) 702–706

Camacho J. and Zakari M., Irreversible thermodynamic analysis of a two-layer system,

Phys. Rev. E 50 (1994) 4233–4236

Caplan S.R. and Essig A., Bioenergetics and Linear Nonequilibrium Thermodynam-

ics. The Steady State, Harvard University Press, Cambridge, 1983

Carrassi M. and Morro A., A modiﬁed Navier–Stokes equations and its consequences

on sound dispersion, Nuovo Cimento B 9 (1972) 321–343; 13 (1984) 281–289

Casas-V´azquez J. and Jou D., Temperature in non-equilibrium states: a review of

open problems and current proposals, Rep. Prog. Phys. 66 (2003) 1937–2023

Casas-V´azquez J., Jou D. and Lebon G. (eds.), Recent Developments in Non-

equilibrium Thermodynamics, Lecture Notes in Physics, vol. 199, Springer, Berlin

Heidelberg New York, 1984

Casimir H.B.G., On Onsager’s principle of microscopic irreversibility, Rev. Mod. Phys.

17 (1945) 343–350

Cattaneo C., Sulla conduzione del calore, Atti Seminario Mat. Fis. University Modena

3 (1948) 83–101

Chandrasekhar S., Hydrodynamic and Hydromagnetic Stability, Clarendon, Oxford,

1961

Chapman S. and Cowling T.G., The Mathematical Theory of Non-Uniform Gases,

Cambridge University Press, Cambridge, 1970

Chartier P., Gross M. and Spiegler K.S., Applications de la thermodynamique du

non-´equilibre, Hermann, Paris, 1975 (in French)

Chen G., Nanoscale Heat Transfer and Conversion, Oxford University Press, Oxford,

2004

References 309

Cimmelli V.A., Boundary conditions in the presence of internal variables, J. Non-

Equilib. Thermodyn. 27 (2002) 327–348

Cloot A. and Lebon G., A nonlinear stability analysis of the B´enard–Marangoni

problem, J. Fluid Mech. 45 (1984) 447–469

Coleman B.D., Thermodynamics of materials with memory, Arch. Rat. Mech. Anal.

17 (1964) 1–46

Coleman B.D. and Owen D., A mathematical foundation of thermodynamics, Arch.

Rat. Mech. Anal. 54 (1974) 1–104

Coleman B.D. and Owen D., On thermodynamics and elastic–plastic materials, Arch.

Rat. Mech. Anal. 59 (1975) 25–51

Coleman B.D. and Truesdell C., On the reciprocal relations of Onsager, J. Chem.

Phys. 33 (1960) 28–31

Coleman B.D., Owen D. and Serrin J., The second law of thermodynamics for systems

with approximate cycles, Arch. Rat. Mech. Anal. 77 (1981) 103–142

Coles D., Transition in circular Couette ﬂow, J. Fluid Mech. 21 (1965) 385–425

Colinet P., Legros J.C. and Velarde M., Nonlinear Dynamics of Surface-Tension

Driven Instabilities, Wiley, Berlin, 2002

Criado-Sancho M. and Casas-V´azquez J., Termodin´ amica qu´ımica y de los procesos

irreversibles, 2nd ed., Pearson/Addison-Wesley, Madrid, 2004 (in Spanish)

Criado-Sancho M. and Llebot J.E., Behaviour of entropy in hyperbolic heat transport,

Phys. Rev. E 47 (1993) 4104–4108

Crisanti A. and Ritort F., Violation of the ﬂuctuation–dissipation theorem in glassy

systems: basic notions and the numerical evidence, J. Phys. A: Math. Gen. 36

(2003) R181–R290

Cross M.C. and Hohenberg P.C., Pattern formation outside equilibrium, Rev. Mod.

Phys. 65 (1993) 852–1112

Curzon F.L. and Ahlborn B., Eﬃciency of a Carnot engine at maximum power output,

Am. J. Phys. 43 (1975) 22–24

Dauby P.C. and Lebon G., B´enard–Marangoni instability in rigid rectangular con-

tainers, J. Fluid Mech. 329 (1996) 25–64

Dauby P.C., Bouhy E. and Lebon G., Linear B´enard–Marangoni instability in rigid

circular containers, Phys. Rev. E 56 (1997) 520–530

Davis S.H., Buoyancy-surface tension instability by the method of energy, J. Fluid

Mech. 39 (1969) 347–359

Davis S.H., Thermocapillary instabilities, Annu. Rev. Fluid Mech. 19 (1987) 403–435

Day W., The Thermodynamics of Simple Materials with Fading Memory, Springer,

Berlin Heidelberg New York, 1972

Day W., An objection against using entropy as a primitive concept in continuum

thermodynamics, Acta Mechanica 27 (1977) 251–255

Dedeurwaerdere T., Casas-V´azquez J., Jou D. and Lebon G., Foundations and appli-

cations of a mesoscopic thermodynamic theory of fast phenomena, Phys. Rev. E

53 (1996) 498–506

Doi M. and Edwards S.F., The Theory of Polymer Dynamics, Clarendon, Oxford,

1986

Drazin P.G. and Reid W.H., Hydrodynamic Stability, Cambridge University Press,

Cambridge, 1981

Dreyer W. and Struchtrup H., Heat pulse experiments revisited, Continuum Mech.

Thermodyn. 5 (1993) 3–50

Duhem P., Trait´e d’´energ´etique ou de thermodynamique g´en´erale, Gauthier-Villars,

Paris, 1911

Eckart C., The thermodynamics of irreversible processes. I–III, Phys. Rev. 58 (1940)

267–269, 269–275, 919–924; IV, Phys. Rev. 73 (1948) 373–382

Edelen D.G.D. and McLennan J.A., Material indiﬀerence: a principle or a conve-

nience, Int. J. Eng. Sci. 11 (1973) 813–817

310 References

Elmer F.J., Limit cycle of the ballast resistor caused by intrinsic instabilities, Z. Phys.

B 87 (1992) 377–381

Eringen C., Mechanics of Continua, Wiley, New York, 1967

Eu B.C., Kinetic Theory and Irreversible Thermodynamics, Wiley, New York, 1992

Eu B.C., Nonequilibrium Statistical Mechanics: Ensemble Method, Kluwer,

Dordrecht, 1998

Evans D.J. and Morriss G.P., Statistical Mechanics of Non-Equilibrium Liquids,

Academic, London, 1990

Finlayson B., The Method of Weighted Residuals and Variational Principles,

Academic, New York, 1972

Fowler R.H. and Guggenheim E.A., Statistical Thermodynamics, Cambridge

University Press, Cambridge, 1939

Frankel N.A. and Acrivos A., The constitutive equations for a dilute emulsion,

J. Fluid Mech. 44 (1970) 65–78

Garc´ıa-Colin L.S., Extended irreversible thermodynamics beyond the linear regime:

a critical overview, J. Non-Equilib. Thermodyn. 16 (1991) 89–128

Garcia-Colin L.S., Extended irreversible thermodynamics: an unﬁnished task, Mol.

Phys. 86 (1995) 697–706

Gaspard P., Chaos, Scattering and Statistical Mechanics, Cambridge University

Press, Cambridge, 1998

Gibbs J.W., The Collected Works of J.W. Gibbs, vol. 1: Thermodynamics, Yale

University Press, Yale, 1948

Glansdorﬀ P. and Prigogine I., On a general evolution criterion in macroscopic

physics, Physica 30 (1964) 351–374

Glansdorﬀ P. and Prigogine I., Thermodynamics of Structure, Stability and Fluctu-

ations, Wiley, New York, 1971

Gollub J.P. and Langer J.S., Pattern formation in nonequilibrium physics, Rev. Mod.

Phys. 71 (1999) S396–S403

Gorban A.N., Karlin I.V. and Zinovyev A.Y., Constructive methods of invariant

manifolds for kinetic problems, Phys. Rep. 396 (2004) 197–403

Gordon J.M. and Zarmi Y., Wind energy a solar-driven heat engine: a thermodynamic

approach, Am. J. Phys. 57 (1989) 995–998

Grad H., Principles of the Kinetic Theory of Gases, in Hd. der Physik, vol. XII (Flugge

S., ed.), Springer, Berlin Heidelberg New York, 1958

Grandy W.T., Foundations of Statistical Mechanics, Reidel, Dordrecht, 1987

Grmela M., Complex ﬂuids subjected to external inﬂuences, J. Non-Newtonian Fluid

Mech. 96 (2001) 221–254

Grmela M. and Lebon G., Hamiltonian extended thermodynamics, J. Phys. A: Math.

Gen. 23 (1990) 3341–3351

Grmela M. and

¨

Ottinger H.C., Dynamics and thermodynamics of complex ﬂuids. I.

Development of a generic formalism, Phys. Rev. E 56 (1997) 6620–6632

Grmela M., Elaﬁf A. and Lebon G., Isothermal non-standard diﬀusion in a two-

component ﬂuid mixture: a Hamiltonian approach, J. Non-Equilib. Thermodyn.

23 (1998) 312–327

Grmela M., Lebon G. and Lhuillier D., A comparative study of the coupling of ﬂow

with non-Fickean thermodiﬀusion. II. Generic, J. Non-Equilib. Thermodyn. 28

(2003) 23–50

Grmela M., Lebon G., Dauby P.C. and Bousmina M., Ballistic-diﬀusive heat conduc-

tion at nanoscale: a generic approach, Phys. Lett. A 339 (2005) 237–245

de Groot S.R. and Mazur P., Non-Equilibrium Thermodynamics, North-Holland,

Amsterdam, 1962 (republication by Dover in 1984)

Guyer R.A. and Krumhansl J.A., Solution of the linearized Boltzmann phonon equa-

tion, Phys. Rev. 148 (1966) 766–788

References 311

Gyarmati I., Non-equilibrium Thermodynamics, Springer, Berlin Heidelberg New

York, 1970

Hafskold B. and Kjelstrup S., Criteria for local equilibrium in a system with transport

of heat and mass, J. Stat. Phys. 78 (1995) 463–494

Haken H. (ed.), Synergetics: A Workshop, Springer, Berlin Heidelberg New York,

1977

H¨ ansch W., The Drift-Diﬀusion Equation and Its Application in MOSFET Modeling,

Springer, Berlin Heidelberg New York, 1991

Hill T.L., Free Energy Transduction in Biology, the Steady-State Kinetic and Ther-

modynamic Formalism, Academic, New York, 1977

Hoﬀmann K.H., Burzler J.M. and Schubert S., Endoreversible thermodynamics,

J. Non-Equilib. Thermodyn. 22 (1997) 311–355

Hoover W.H., Time Reversibility, Computer Simulations and Chaos, World Scientiﬁc,

Singapore, 1999

Hoover W., Moran B., More R. and Ladd A., Heat conduction in a rotating disk via

non-equilibrium molecular dynamics, Phys. Rev. A 24 (1981) 2109–2115

Hutter K. and J¨ohnk K., Continuum Methods of Physical Modelling, Springer, Berlin

Heidelberg New York, 2004

Ichiyanagi M., Variational principles in irreversible processes, Phys. Rep. 243 (1994)

125–182

Ilg P. and

¨

Ottinger H.C., Non-equilibrium relativistic thermodynamics in bulk viscous

cosmology, Phys. Rev. D 61 (1999) 023510, 1–10

Jaynes E.T., in Statistical Physics (Ford W.K., ed.) Benjamin, New York, 1963

Joseph D.D., Stability of Fluid Motions, Springer, Berlin Heidelberg New York, 1976

Joseph D.D. and Preziosi L., Heat waves, Rev. Mod. Phys. 61 (1989) 41–74; 62 (1990)

375–392

Jou D. and Llebot J.E., Introduction to the Thermodynamics of Biological Processes,

Prentice Hall, New York, 1990

Jou D., Casas-V´azquez J. and Criado-Sancho M., Thermodynamics of Fluids Under

Flow, Springer, Berlin Heidelberg New York, 2000

Jou D., Casas-V´azquez J. and Lebon G., Extended Irreversible Thermodynamics, 3rd

ed., Springer, Berlin Heidelberg New York, 2001

Jou D., Casas-V´azquez J., Lebon G. and Grmela M., A phenomenological scaling

approach for heat transport in nano-systems, Appl. Math. Lett. 18 (2005) 963–

967

J¨ ulicher A., Ajdari A. and Prost P., Modeling molecular motors, Rev. Mod. Phys. 69

(1997) 1269–1282

Katchalsky A. and Curran P.F., Non-Equilibrium Thermodynamics in Biophysics,

Harvard University Press, Cambridge, 1965

Keizer J., Thermodynamics of non-equilibrium steady states, J. Chem. Phys. 69

(1978) 2609–2620

Keizer J, On the relationship between ﬂuctuating irreversible thermodynamics and

‘extended’ irreversible thermodynamics, J. Stat. Phys. 31 (1983) 485–497

Keizer J., Statistical Thermodynamics of Nonequilibrium Processes, Springer, Berlin

Heidelberg New York, 1987

Kestin J., A Course in Thermodynamics, Blaisdell, Waltham, 1968

Kestin J., A note on the relation between the hypothesis of local equilibrium and the

Clausius–Duhem inequality, J. Non-Equilib. Thermodyn. 15 (1990) 193–212

Kestin J., Local-equilibrium formalism applied to mechanics of solids, Int. J. Solids

Struct. 29 (1992) 1827–1836

Kestin J. and Bataille J., Thermodynamics of Solids, in Continuum Models of Dis-

crete Systems, University of Waterloo Press, Waterloo, 1980

Khalatnikov I.M., An introduction to the Theory of Superﬂuidity, W.A. Benjamin,

New York, 1965

312 References

Kleidon A. and Lorenz R.D. (eds.), Non-Equilibrium Thermodynamics and the

Production of Entropy. Life, Earths and Beyond, Springer, Berlin Heidelberg New

York, 2005

Kluitenberg G., Thermodynamical theory of elasticity and plasticity, Physica 28

(1962) 217–232

Kondepudi D. and Prigogine I., Modern Thermodynamics: from Heat Engines to

Dissipative Structures, Wiley, New York, 1998

Koschmieder E.L., B´enard cells and Taylor Vortices, Cambridge University Press,

Cambridge, 1993

Kratochvil J. and Silhavy M., On thermodynamics of non-equilibrium processes,

J. Non-Equilib. Thermodyn. 7 (1982) 339–354

Kreuzer H.J., Non-Equilibrium Thermodynamics and Its Statistical Foundations,

Clarendon, Oxford, 1981

Kubo R., Statistical Mechanics, Wiley, New York, 1965

Landau L.D., Collected Papers (Haar T., ed.), Pergamon, Oxford, 1965

Landau L.D. and Lifshitz E.M., Statistical Physics, 3rd ed., Pergamon, Oxford, 1980

Landsberg P.T. and Tonge G., Thermodynamic energy conversion eﬃciencies, J. Appl.

Phys. 51 (1980) R1–R18

Lavenda B., Thermodynamics of Irreversible Processes, McMillan, London, 1979

Lebon G., Variational Principles in Thermomechanics, in CISME Courses and Lec-

ture Notes, vol. 262, Springer, Berlin Heidelberg New York, 1980

Lebon G. and Jou D., Linear irreversible thermodynamics and the phenomenological

theory of liquid helium II, J. Phys. C 16 (1983) L199–L204

Lebon G. and Dauby P.C., Heat transport in dielectric crystals at low temperature:

a variational formulation based on extended irreversible thermodynamics, Phys.

Rev. A 42 (1990) 4710–4715

Lebon G., Extended Thermodynamics, in CISME Courses and Lecture Notes, vol.

336, pp 139–204, Springer, Berlin Heidelberg New York, 1992

Lebon G., Casas-Vazquez J. and Jou D., Questions and answers about a thermody-

namic theory of the third type, Contemp. Phys. 33 (1992) 41–51

Lebon G., Torrissi M. and Valenti A., A nonlocal thermodynamic analysis of second

sound propagation in crystalline dielectrics, J. Phys. C 7 (1995) 1461–1474

Lebon G., Grmela M. and Lhuillier D., A comparative study of the coupling of

ﬂow with non-Fickian thermodiﬀusion. I. Extended irreversible thermodynamics,

J. Non-Equilib. Thermodyn. 28 (2003) 1–22

Lebon G., Desaive Th. and Dauby P.C., A uniﬁed extended thermodynamic descrip-

tion of diﬀusion, thermo-diﬀusion, suspensions, and porous media, J. Appl. Mech.

73 (2006a) 16–20

Lebon G., Desaive Th. and Dauby P.C., Two-ﬂuid diﬀusion as a model for heat

transport at micro and nanoscales, in Proceedings of 7th International Meeting on

Thermodiﬀusion, San Sebastian, 2006b

Lebon G., Lhuillier D. and Palumbo A., A thermodynamic description of thermodif-

fusion in suspensions of rigid particles, Eur. Phys. J.IV 146 (2007) 3–12

Lebowitz J.L., Microscopic origins of irreversible macroscopic behaviour, Physica A

263 (1999) 516–527

Lepri S., Livi R. and Politi A., Thermal conduction in classical low-dimensional lat-

tice, Phys. Rep. 377 (2003) 1–80

Lhuillier D., From molecular mixtures to suspensions of particles, J. Phys. II France

5 (1995) 19–36

Lhuillier D., Internal variables and the non-equilibrium thermodynamics of colloidal

suspensions, J. Non-Newtonian Fluid Mech. 96 (2001) 19–30

Lhuillier D. and Ouibrahim A., A thermodynamic model for solutions of deformable

molecules, J. M´ecanique 19 (1980) 725–741

References 313

Lhuillier D., Grmela M. and Lebon G., A comparative study of the coupling of

ﬂow with non-Fickian thermodiﬀusion. Part III. internal variables, J. Non-Equilib.

Thermodyn. 28 (2003) 51–68

Liu I.-S., Method of Lagrange multipliers for exploitation of the entropy principle,

Arch. Rat. Mech. Anal. 46 (1972) 131–148

Lorenz E.N., Deterministic nonperiodic ﬂow, J. Atmos. Sci. 20 (1963) 130–141

Lumley J., Turbulence modelling, J. Appl. Mech. 50 (1983) 1097–1103

Luzzi R., Vasconcellos A.R. and Ramos J.G., Statistical Foundations of Irreversible

Thermodynamics, Teubner, Leipzig, 2001

Luzzi R., Vasconcellos A.R. and Ramos J.G., Predictive Statistical Mechanics: A

Non-equilibrium Ensemble Formalism, Kluwer, Dordrecht, 2002

Lyapounov M., Stability of Motion, Academic, New York, 1966 (English translation)

Ma S.K., Statistical Mechanics, World Scientiﬁc, Singapore, 1985

Madruga S., P´erez-Garc´ıa C. and Lebon G., Convective Instabilities in two laterally

superposed horizontal layers heated, Phys. Rev. E 68 (2003) 041607, 1–12

Magnasco M.O., Molecular combustion motors, Phys. Rev. Lett. 72 (1994) 2656–2659

Mahan G., Sales B. and Sharp J., Thermoelectric materials: new approaches to an

old problem, Phys. Today 50 (1997) 42–47

Mandel J., Propri´et´es M´ecaniques des Mat´eriaux, Eyrolles, Paris, 1978

Maugin G., The Thermodynamics of Nonlinear Irreversible Behaviors, World Scien-

tiﬁc, Singapore, 1999

Maugin G. and Muschik W., Thermodynamics with internal variables, J. Non-Equilib.

Thermodyn. 19 (1994) 217–289

Maurer M.J. and Thomson H.A., Non-Fourier eﬀects at high heat ﬂux, J. Heat Trans-

fer 95 (1973) 284–286

Maxwell J.C., On the dynamical theory of gases, Philos. Trans. R. Soc. Lond. 157

(1867) 49–88

McDougall J.T. and Turner J.S., Inﬂuence of cross-diﬀusion on ‘ﬁnger’ double-

diﬀusive convection, Nature 299 (1982) 812–814

McQuarrie D.A., Statistical Mechanics, Harper and Row, New York, 1976

Meixner J., Zur thermodynamik der irreversiblen prozesse in gasen mit chemisch

reagierenden, dissozierenden und anregbaren komponenten, Ann. Phys. (Leipzig)

43 (1943) 244–269

Meixner J., The entropy problem in thermodynamics of processes, Rheol. Acta 12

(1973a) 272–283

Meixner J., On the Foundations of Thermodynamics of Processes, in A Critical Re-

view of Thermodynamics (Stuart E., Gal-Or B. and Brainard A., eds.), Mono

Book Corp., Baltimore, 1973b

Meixner J. and Reik H.G., Thermodynamik der Irreversible Prozesse, in Handbuch

der Physik, Bd 3/ II, Springer, Berlin Heidelberg New York, 1959

Mongiov`ı M.S., Extended irreversible thermodynamics of liquid helium II, Phys. Rev.

B 48 (1993), 6276–6289

Mongiov`ı M. S., Proposed measurements of the small entropy carried by the superﬂuid

component of liquid helium II, Phys. Rev. B. 63 (2001) 12501–12510

Morse P.M. and Feshbach H., Mathematical Methods of Theoretical Physics, McGraw

Hill, New York, 1953

Movchan A., The direct method of Lyapounov in stability problems of elastic systems,

J. Appl. Math. Mech. 23 (1959) 686–700

M¨ uller I., On the frame dependence of stress and heat ﬂux, Arch. Rat. Mech. Anal.

45 (1972) 241

M¨ uller I. and Ruggeri T., Rational Extended Thermodynamics, 2nd ed., Springer,

Berlin Heidelberg New York, 1998

Muschik W., Empirical foundation and axiomatic treatment of non-equilibrium tem-

perature, Arch. Rat. Mech. Anal. 66 (1977) 379–401

314 References

Muschik W., Aspects of Non-Equilibrium Thermodynamics, World Scientiﬁc,

Singapore, 1990

Nelson P., Biological Physics. Energy, Information, Life, Freeman, New York, 2004

Nicolis G. and Nicolis C., On the entropy balance of the earth-atmosphere system,

Q. J. R. Meteorol. Soc. 106 (1980) 691–706

Nicolis G. and Prigogine I., Self-Organization in Nonequilibrium Systems, Wiley, New

York, 1977

Nicolis G. and Prigogine I., Exploring Complexity, W.H. Freeman, New York, 1989

Nield D.A., Surface tension and buoyancy eﬀects in cellular convection, J. Fluid Mech.

19 (1964) 341–353

Noll W., The Foundations of Mechanics and Thermodynamics, Springer, Berlin, 1974

Nettleton R.E. and Sobolev S.L., Applications of extended thermodynamics to

chemical, rheological and transport processes: a special survey, J. Non-Equilib.

Thermodyn. 20 (1995) 200–229, 297–331

Onsager L., Reciprocal relations in irreversible processes, Phys. Rev. 37 (1931) 405–

426; 38 (1931) 2265–2279

Onuki A., Phase transitions of ﬂuids in shear ﬂow, J. Phys.: Condens. Mat. 9 (1997)

6119–6157

Onuki A., Phase Transitions Dynamics, Cambridge University Press, Cambridge,

2002

¨

Ottinger H.C., A thermodynamically admissible reptation model for fast ﬂows of

entangled polymers, J. Rheol. 43 (1999a) 1461–1493

¨

Ottinger H.C., Thermodynamically admissible equations for causal dissipative cos-

mology, galaxy formation, and transport processes in gravitational collapse, Phys.

Rev. D 60 (1999b) 103507, 1–9

¨

Ottinger H.C., Beyond Equilibrium Thermodynamics, Wiley, Hoboken, 2005

¨

Ottinger H.C., Non-equilibrium thermodynamics of open systems, Phys. Rev. E 73

(2006) 036126

¨

Ottinger H.C. and Grmela M., Dynamics and thermodynamics of complex ﬂuids. II.

Illustrations of a general formalism, Phys. Rev. E 56 (1997) 6633–6655

Ozawa H., Ohmura A., Lorenz R.D. and Pujol T., The second law of thermodynam-

ics and the global climate system: a review of the maximum entropy production

system, Rev. Geophys. 41 (2003) 4/1018

Palm E., Nonlinear thermal convection, Annu. Rev. Fluid Mech. 7 (1975) 39–61

Paltridge G.W., Thermodynamic dissipation and the global climate system, Q. J. R.

Meteorol. Soc. 107 (1981) 531–547

Parmentier P., Regnier V. and Lebon G., Buoyant-thermocapillary instabilities in

medium Pr number ﬂuid layers subject to a horizontal temperature gradient, Int.

J. Heat Mass Transfer 36 (1993) 2417–2427

Pasmanter R.A., Bedeaux D. and Mazur P., The ballast resistor: an electro-thermal

instability in a conducting wire II; ﬂuctuations around homogeneous stationary

states, Physica A 90 (1978) 151–163

Pearson J.R.A., On convection cells induced by surface tension, J. Fluid Mech. 4

(1958) 489–500

Peixoto J.P. and Oort A.H., The physics of climate, Rev. Mod. Phys. 56 (1984)

365–429

Ponter A.S., Bataille J. and Kestin J., A thermodynamic model for the time inde-

pendent plastic deformation of solids, J. M´ecanique 18 (1978) 511–539

Prigogine I., Etude thermodynamique des ph´enom`enes irr´eversibles, Desoer, Li`ege,

1947

Prigogine I., Introduction to Thermodynamics of Irreversible Processes, Interscience,

New York, 1961

Prigogine I., La ﬁn des certitudes, O. Jacob, Paris, 1996

References 315

Prigogine I. and Lefever R., Symmetry breaking instabilities in dissipative systems,

J. Chem. Phys. 48 (1968) 1695–1700

Prigogine I. and Stengers I., La nouvelle alliance, Gallimard, Paris, 1979

Pritchard A.J., A study of the classical problem of hydrodynamic stability, J. Inst.

Math. Applics. 4 (1968) 78–93

Pujol T. and Llebot J.E., Extremal principle of entropy production in the climate

system, Q. J. R. Meteorol. Soc. 125 (1999) 79–90

Rayleigh Lord, Scientiﬁc Papers, vol. VI, Cambridge University Press, Cambridge,

1920

Rebhan E., Eﬃciency of non-ideal Carnot engines with friction and heat losses, Am.

J. Phys. 70 (2002) 1143–1149

Regnier V., Dauby P.C. and Lebon G., Linear and nonlinear Rayleigh–B´enard–

Marangoni instability with surface deformations, Phys. Fluids 12 (2000) 2787–

2799

Reguera D., Mesoscopic nonequilibrium kinetics of nucleation processes, J. Non-

Equilib. Thermodyn. 29 (2004) 327–344

Reguera D., Rub´ı J.M. and Vilar J.M., The mesoscopic dynamics of thermodynamic

systems, J. Phys. Chem. B 109 (2005) 21502–21515

Reichl L.E., A Modern Course in Statistical Physics, University Texas Press, Austin,

1998

Rivlin R., Forty Years of Non-Linear Continuum Mechanics, in Proceedings of the 9th

International Congress on Rheology, Mexico, 1984

Rosenblat S., Davis S.H. and Homsy G.M., Nonlinear Marangoni convection in

bounded layers, J. Fluid Mech. 120 (1982) 91–138

Rub´ı J.M., The non-equilibrium thermodynamics approach to the dynamics of meso-

scopic systems, J. Non-Equilib. Thermodyn. 29 (2004) 315–325

Rub´ı J.M. and Casas-V´azquez J., Thermodynamical aspects of micropolar ﬂuids. A

non-linear approach, J. Non-Equilib. Thermodyn. 5 (1980) 155–164

Ruelle D., Chance and Chaos, Princeton University Press, Princeton, 1991

Salamon P. and Sieniutycz S. (eds.), Finite-Time Thermodynamics and Optimization,

Taylor and Francis, New York, 1991

Salamon P., Nitzan A., Andresen B. and Berry R.S., Minimum entropy production

and the optimization of heat engines, Phys. Rev. A 21 (1980) 2115–2129

Scriven L.E. and Sternling C.V., On cellular convection driven by surface tension

gradients: eﬀect of mean surface tension and viscosity, J. Fluid Mech. 19 (1964)

321–340

Segel L., Non-linear hydrodynamic stability theory and its application to thermal

convection and curved ﬂows, in Non-Equilibrium Thermodynamics, Variational

Techniques and Stability (Donelly R.F., Herman R. and Prigogine I., eds.), pp

165–197, University of Chicago Press, Chicago, 1966

Serrin J., Conceptual analysis of the classical second law of thermodynamics, Arch.

Rat. Mech. Anal. 70 (1979) 355–371

Shannon C., A mathematical theory of communication, Bell System Tech. J. 27

(1948) 379–423, 623–656

Shl¨ uter A., Lortz D. and Busse F.H., On the stability of steady ﬁnite amplitude

convection, J. Fluid Mech. 23 (1965) 129–144

Sidoroﬀ F.J., Variables internes en visco´elasticit´e, J. M´ecanique 14 (1975) 545–566,

571–595; 15 (1976) 95–118

Sieniutycz S., Conservation Laws in Variational Thermohydrodynamics, Kluwer,

Dordrecht, 1994

Sieniutycz S. and Farkas H. (eds.), Variational and Extremum Principles in Macro-

scopic Systems, Elsevier Science, Oxford, 2004

Sieniutycz S. and Salamon P. (eds.), Extended Thermodynamic Systems, Taylor and

Francis, New York, 1992

316 References

Silhavy M., The Mechanics and Thermodynamics of Continuous Media, Springer,

Berlin Heidelberg New York, 1997

Snider R.F. and Lewchuk K.S., Irreversible thermodynamics of a ﬂuid system with

spin, J. Chem. Phys. 46 (1967) 3163–3172

Snyder G.J. and Ursell T.S., Thermoelectric eﬃciency and compatibility, Phys. Rev.

Lett. 91 (2003) 148301

Sparrow C., The Lorenz Equations: Bifurcations, Chaos and Strange Attractors,

Springer, Berlin Heidelberg New York, 1982

Spencer A.J.M. and Rivlin R.S., Further results in the theory of matrix polynomials,

Arch. Rat. Mech. Anal. 4 (1959) 214–230

Stuart J.T., On the non-linear mechanics of hydrodynamic instability, J. Fluid Mech.

4 (1958) 1–21

Swift J. and Hohenberg P.C., Hydrodynamic ﬂuctuation at the convective instability,

Phys. Rev. A 15 (1977) 319–328

Thom R., Stabilit´e structurelle et morphog´en`ese, Benjam´ın, Reading, MA, 1972

Tisza L. and Manning I., Fluctuations and irreversible thermodynamics, Phys. Rev.

105 (1957) 1695–1704

Truesdell C., Thermodynamics for Beginners, in Irreversible Aspects of Continuum

Mechanics (Parkus H. and Sedov L., eds.), Springer, Berlin Heidelberg New York,

1968

Truesdell C., Rational Thermodynamics, 2nd ed., Springer, Berlin Heidelberg New

York, 1984

Truesdell C. and Noll W., The Non-Linear Field Theories, in Handbuch der Physik,

Bd. III/3, Springer, Berlin Heidelberg New York, 1965

Truesdell C. and Toupin R., The Classical Field Theories, in Handbuch der Physik,

Bd. III/1, Springer, Berlin Heidelberg New York, 1960

Tsallis C., Some thoughts on theoretical physics, Physica A 344 (2004) 718–736

Turcotte D.L., Fractals and Chaos in Geology and Geophysics, Cambridge University

Press, Cambridge, 1992

Turing A., The chemical basis of morphogenesis, Philos. Trans. R. Soc. Lond. B 237

(1952) 37–72

Tzou D.Y., Macro-to-Microscale Heat Transfer. The Lagging Behaviour, Taylor and

Francis, New York, 1997

Valanis K.C., A gradient theory of internal variables, Acta Mechanica 116 (1996)

1–14

Valenti A., Torrisi M. and Lebon G., Shock waves in crystalline dielectrics at low

temperature, J. Phys. Condens. Matter 14 (2002) 3553–3564

Van den Broeck C., Taylor dispersion revisited, Physica A 168 (1990) 677–696

Van den Broeck C., Thermodynamic eﬃciency at maximum power, Phys. Rev. Lett.

95 (2005) 190602

Van Kampen N.G., Views of a Physicist, World Scientiﬁc, Singapore, 2000

Vernotte P., La veritable equation de la chaleur, Compt. Rend. Acad. Sci. Paris 247

(1958) 2103–2107

Veronis G., Cellular convection with a ﬁnite amplitude in a rotating ﬂuid, J. Fluid

Mech. 5 (1959) 401–435

Vidal C., Dewel G. and Borckmans P., Au del` a de l’´equilibre, Hermann, Paris, 1994

Vilar J.M.G. and Rub´ı J.M., Thermodynamics beyond local equilibrium, Proc. Natl

Acad. Sci. USA 98 (2001) 11081–11084

de Vos A., Endoreversible Thermodynamics of Solar Energy Conversion, Oxford Uni-

versity Press, Oxford, 1992

Westerhoﬀ N.V. and van Dam K., Thermodynamics and Control of Biological Free-

Energy Transduction, Elsevier, Amsterdam, 1987

Wilmanski K., Thermodynamics of Continua, Springer, Berlin Heidelberg New York,

1998

References 317

Woods L., The Thermodynamics of Fluid Systems, Clarendon, Oxford, 1975

Woods L., The bogus axioms of continuum mechanics, Bull. Inst. Math. Appl. 17

(1981) 98–102; 18 (1982) 64–67

Zemansky M.W., Heat and Thermodynamics, 5th ed., McGraw-Hill, New York, 1968

Further Readings

There exists a multiplicity of textbooks and popular books on the subject and it would

be unrealistic to mention all of them. The list given below is therefore limited and,

of course, subjective.

Atkins P.W., The Second Law, Scientiﬁc American Library, W.H. Freeman, New York,

1984

Bailyn M., A Survey of Thermodynamics, AIP, New York, 1993

Coveney P. and Highﬁeld R., The Arrow of Time, Flamingo, London, 1990

Davies P.C.W., The Physics of Time Asymmetry, University of California Press,

Berkeley, 1974

Denbigh K. and Denbigh J., Entropy in Relation with Incomplete Knowledge,

Cambridge University Press, Cambridge, 1985

Feynman R.P. and Weinberg S., Elementary Particles and the Laws of Physics,

Cambridge University Press, Cambridge, 1987

Feynman R.P., Leighton R.B. and Sands M., The Feynman Lectures on Physics,

Addison-Wesley, 1963 (Chaps. 39–46)

Goodwin B., How the Leopard Changed Its Spots: The Evolution of Complexity,

Weidenfeld and Nicolson, London, 1994

Harman P.M., Energy, Force and Matter: The Conceptual Development of

Nineteenth-Century Physics, Cambridge University Press, Cambridge, 1982

Kestin J., A Course in Thermodynamics, 2 vols., Hemisphere, Washington, 1979

Longair M.S., Theoretical Concepts in Physics: An Alternative Views of Theoretical

Reasoning in Physics, 2nd ed., Cambridge University Press, Cambridge, 2003

Maxwell J.C., Theory of Heat, Dover, New York, 2001

Pippard A.B., The Elements of Classical Thermodynamics, Cambridge University

Press, London, 1957

Prigogine I., From Being to Becoming, Freeman, San Francisco, 1980

Prigogine I. and Stengers I., Entre le temps et l’´eternit´e, Fayard, Paris, 1988

Segr`e G., Einstein’s Refrigerator, Penguin Books, London, 2004

Tisza L. and Manning T., Fluctuations and irreversible thermodynamics, Phys. Rev.

105 (1957) 1695–1704

Vidal Ch. and Lemarchand H., La r´eaction cr´eatrice, Herman, Paris, 1988

Walgraef D., Spatio-Temporal Pattern Formation, Springer, Berlin Heidelberg New

York, 1997

Winfree A., The Geometry of Biological Time, Springer, Berlin Heidelberg New York,

1980

Winfree A., The Timing of Biological Clocks, Scientiﬁc American Library, Freeman,

New York, 1987a

Winfree A., When Time Breaks Down, Princeton University Press, Princeton, 1987b

319

Index

absolute temperature, 2, 9, 10, 14, 40,

233, 250

accompanying state, 216, 217, 220, 222,

225, 233

active transport, 83, 91, 103, 106, 111

adiabatic, 217

adiabatic wall, 5, 6, 18

aﬃnity, 30, 32, 33, 47, 93, 94, 104, 106,

110, 219, 222, 297

ageing, 44

albedo, 125–128, 133

amplitude method, 136, 139, 156

angular velocity, 49, 67, 138, 158, 159,

161, 175, 242, 243, 251, 276, 285

anisotropic systems, 39, 65, 138

arrow of time, 43

asymptotically stable, 135, 164

atmosphere, 82, 125, 126, 128, 132, 144

autocatalytic reaction, 108, 111, 163,

169, 177

availability, 124, 280

balance equations, 45, 53, 59, 60, 80, 94,

143, 170, 203, 216, 219, 225, 238,

239, 243, 244, 247, 250, 253, 261,

275, 276

ballast resistor, 171, 173

ballistic regime, 182, 191, 192, 195

barycentric velocity, 45, 59, 79, 224, 278

B´enard–Marangoni’s instability, 145,

158, 176

Belousov–Zhabotinsky reaction, 166

bifurcation, 135, 141, 157, 162, 171

biology, 38, 44, 54, 91, 100, 107, 109,

162, 169, 305

black body, 132

Boltzmann–Planck’s formula, 280

boundary conditions, 51–54, 57, 58, 144,

148, 149, 151, 152, 155, 159, 160,

175, 176, 208, 219, 250, 254, 268

Boussinesq approximation, 147, 158

Brownian motion, 106, 107, 235, 285,

293, 298, 300

Brownian Motors, 106

Brusselator models, 163

bulk viscosity, 62, 199, 248, 258

buoyancy–surface-tension, 177

Carnot cycle, 10, 34, 114

catalyses, 107

Cattaneo equation, 191–193, 211, 240

causality, 58

chaos, 44, 136, 153, 303

chemical instabilities, 162, 172

chemical kinetics, 91, 99, 166

chemical potential, 4, 14–16, 19, 28, 30,

40, 59, 70, 80, 83, 84, 93, 202, 203,

213, 225, 231, 268, 278, 290, 291,

293, 297

chemical reactions, 5, 14, 23, 29, 31, 38,

39, 46, 54, 63, 65, 79, 83, 91, 96

classical irreversible thermodynamics,

37, 44, 58, 63, 65, 72, 99, 179,

198, 203, 216, 220, 233, 241, 276,

292–294, 304

Clausius’ inequality, 12, 292

Clausius–Duhem’s inequality, 220, 234,

238, 239, 241, 243, 244, 246, 248,

249, 251, 253, 257–259

climatic changes, 126

clouds, 125, 127

coeﬃcient of thermal expansion, 17, 145,

282

colloidal suspensions, 227, 234

321

322 Index

complex ﬂuids, 224, 226, 227, 275

conﬁguration tensor, 200

conservation laws, 51

constitutive relations, 63, 99, 219, 225,

233, 238, 240–243, 247–249, 251,

253, 272, 304

constrained equilibrium, 218, 221

continued-fraction, 195, 196, 210

control parameter, 135, 138, 139, 163,

174

convected time derivatives, 261

corotational derivative, 261

Couette ﬂow, 158, 162, 177

coupled chemical reactions, 96, 100

coupled processes, 69, 91, 101, 103

critical threshold, 136, 138, 139

critical wave number, 151, 156, 161

Curie’s law, 47, 48

Curzon–Ahlborn’s engine, 129, 130

cyclical chemical reactions, 97

Deborah number, 40, 180, 218

degeneracy conditions, 266, 278

degree of advancement, 29, 92, 95, 96,

234

degree of coupling, 101

dendrites, 172

detailed balance, 97–99, 112

diﬀusion, 18, 32, 38, 51, 54, 57–59, 63

diﬀusion coeﬃcient, 38, 49, 60, 67, 81,

89, 168, 276, 294, 297, 300

diﬀusion ﬂux, 38, 45, 59, 67, 89, 108,

228, 274, 293, 294, 296

dislocations, 216, 223

disorder, 3, 11, 174

dispersion relation, 150, 193, 206

dissipation potential, 265, 266, 269, 270,

272, 274–276

dissipative bracket, 269, 277

dissipative structures, 136

Dufour’s eﬀect, 82

dumbbells, 224

Earth’s Energy Balance, 125

ecology, 53

economy, 91, 205

eﬃciency, 11, 43, 70, 72, 76–78, 88, 91,

92, 100–103

Einstein’s relation, 67

Einstein’s summation convention, 46,

244

Einstein’s theory of ﬂuctuations, 279

electric conductivity, 38

electric current, 38, 47, 73, 75–77, 87,

88, 171, 181, 203, 213, 285

emissivity, 66, 125, 127, 132

empirical temperature, 5, 9, 10

endoreversible heat engine, 117

enthalpy, 7, 8, 21, 23, 24, 27, 31, 80, 232

entropy, 2–4, 8, 10–12, 15, 18, 20, 21, 23

entropy balance, 41, 42, 46, 94, 128, 129,

231

entropy ﬂux, 42, 61, 66, 71, 72, 80, 87,

94, 130, 190, 198, 209, 220, 223,

225, 230, 231, 234, 239, 252, 256,

257, 277

entropy production, 12, 42, 43, 46, 49,

51, 52, 55, 56, 58, 61–63, 65

enzymes, 103

equation of state, 6, 15, 31, 35, 55, 148,

222

equilibrium, 1, 3–9, 11

equilibrium constant, 92, 93

(local) equilibrium, 191, 257, 291

equipresence, 241, 243, 247, 252

Euler’s relation, 16, 21

Eulerian time derivative, 52

evolution equations, 44, 46, 61–63, 70,

79, 136, 180, 189, 192, 194, 197–199

exchange of stability, 138

exothermic reactions, 92

extended irreversible thermodynamics,

58, 179, 188, 189, 200, 202, 207,

215, 232, 233, 240, 252, 276, 278,

286, 304

extensive variables, 4, 15, 19, 217, 220,

288, 290, 291

extremum principles, 21, 51

Fick’s law, 38, 60, 81, 231, 272, 295

ﬁgure of merit, 70, 77, 88

ﬁnite-time thermodynamics, 113, 125,

128

ﬁrst law, 2, 3, 5–8, 11, 14, 19, 22, 31, 92,

114, 117, 187, 217

ﬂuctuation–dissipation theorem, 279,

285, 286

ﬂuctuations, 4, 24–26, 40, 49, 50, 64,

105, 135, 153, 174, 206, 228, 279,

280

ﬂuid ﬂows, 5, 54, 66, 162, 200, 261

ﬂuxes, 4, 38, 39, 45–48, 50, 61, 63, 64,

68, 69, 77, 80, 85, 88, 101, 107

Fokker–Planck’s equation, 236, 295, 297,

298

Fourier transforms, 194

Index 323

Fourier’s law, 38, 56, 65, 66, 68, 76,

181–183, 185, 187, 190, 193, 195,

211, 254, 257

Fourier’s law with, 194

frame-indiﬀerence, 201, 226, 242, 247,

251, 276

friction losses, 120

fundamental relations, 15, 19, 21

generalized hydrodynamics, 64, 212

GENERIC formalism, 266, 274, 277

Gibbs’ equation, 1, 14–16, 32, 40, 45, 59,

62, 63, 70, 72, 80, 189, 191, 197,

203, 213, 219, 222, 225, 230, 234,

235, 238, 244, 248, 249, 256, 281,

290, 293, 298

Gibbs’ free energy, 21, 23, 24, 27, 30, 51,

213

Gibbs–Duhem’s relation, 16, 17, 60, 80

Ginzburg–Landau’s equation, 153

global warming, 125–127, 132

Goldmann equation, 90

Green–Kubo’s formula, 287

greenhouse eﬀect, 126, 127, 132

Guyer–Krumhansl equation, 193, 209,

211

gyroscopic forces, 261, 276

Hall eﬀect, 71

Hamilton equations, 262, 266

Hamiltonian formalisms, 200

harmonic oscillator, 263

heat, 2, 3, 5–9, 11, 14, 18, 19, 21, 26

heat capacity at constant pressure, 17,

282

heat capacity at constant volume, 17, 25

heat conduction, 38, 50, 51, 54, 76, 77,

181, 195, 211, 232, 234, 240, 251,

253, 254, 258, 259, 298, 304

heat conductivity, 49, 55, 56, 58, 62, 63,

65, 72, 77, 81, 149, 170, 176, 181,

191, 194, 196, 232, 245, 248, 256,

257, 287

heat diﬀusivity, 56, 66, 183

heat engines, 10, 88, 101, 103, 113, 114,

123, 131

heat ﬂux, 35, 37, 38, 45, 47, 50, 54, 55,

64, 73, 76, 80, 93, 170, 181

heat of reaction, 31–33

Helmholtz’s free energy, 23, 24, 27, 36,

267, 270, 276, 281

hexagonal patterns, 156

history, 237, 238, 240, 241

Hopf bifurcation, 138

hydrodynamics, 38, 60, 62, 63, 136, 162,

167, 173, 208, 210, 247, 248, 251,

266–268, 270, 275, 277

hyperbolic equations, 184

hysteresis, 143

ideal gas, 13, 30, 31, 34, 35, 93, 114, 117,

195, 287, 299

incompressible ﬂuids, 62

inertial eﬀects, 116, 293, 298

information, 3, 15, 43, 64, 96, 109, 113,

122, 128, 151, 156, 203, 205, 264,

272, 276, 281, 304

instabilities, 44, 135, 136, 145, 158, 162,

170, 173, 212, 279

intensive variables, 4, 15, 16, 19, 24, 41,

46, 52, 65, 80, 290, 292

internal energy, 2, 3, 6, 7, 12, 20, 22, 35,

38–40, 45, 54, 60, 70, 79, 92

internal variables, 179, 200, 215,

217–219, 221, 223

irreversibility, 43, 117, 120, 135

irreversible processes, 5, 12, 15, 24–26,

34, 37, 38, 42–44, 46, 50, 61, 63,

79, 95, 115, 215, 251, 265, 274, 277,

298

isothermal compressibility, 17, 25, 41,

52, 282, 299

isotope separation, 82

isotropic systems, 47, 197

isotropic tensors, 47, 48, 248, 258

Jacobi’s identity, 264, 265, 276, 278

Joule dissipation, 77

Kelvin relation, 73, 76

Kelvin–Planck formulation of the second

law, 114

Kelvin–Voigt model, 223

kinetic constants, 97, 109–111

Knudsen number, 180, 192, 195

Lagrange multipliers, 253, 255

Lagrangian time derivative, 41

Landau’s equation, 141

lasers, 70

Le Chatelier’s principle, 26, 28, 32, 36

Legendre transformations, 19, 24, 27

light scattering, 199, 208, 292

limit cycle, 164, 166

Liu’s Lagrange Multipliers, 253

local equilibrium, 39–41, 45, 63, 179,

185, 186, 189, 191, 197, 202, 207,

324 Index

215–217, 230, 252, 257, 288, 293,

296, 298, 301

Lorenz model, 153, 177

Lotka–Volterra model, 163, 166, 177

Lyapounov function, 53, 58, 270, 290

Lyapounov’s functional, 139

magnetic ﬁelds, 39, 219

Marangoni eﬀect, 144, 177

Marangoni number, 155, 156, 176, 177

marginal stability, 136, 138, 147, 150,

156, 161

mass action law, 110

mass fraction, 40, 45, 59, 60, 70, 80, 81,

92, 96, 228, 271, 273, 278

mass transport, 37, 54, 65, 79, 100, 103,

104, 169, 277, 278

Massieu–Planck functions, 21

materials sciences, 69, 78

Maxwell’s relations, 17, 35

Maxwell–Cattaneo’s equations, 199

mean free path, 70, 192

mechanical ﬁltration coeﬃcient, 85

membranes, 54, 69, 70, 83, 85–87, 89,

102, 292, 297

memory, 179, 181, 206, 207, 219, 237,

240, 241, 247, 251, 257, 286, 287,

300, 304

mesoscopic, 275, 277, 292

metastable, 142

Michaelis–Menten’s, 109, 110

microelectronic devices, 38, 181, 202,

207

micropolar ﬂuids, 67

microscopic reversibility, 49, 64, 284

minimum entropy production, 51, 53,

58, 68, 124, 131, 134

mixtures, 27, 63, 70, 79, 81, 163, 170,

228, 231, 270, 273, 278

mole fraction, 31, 93

molecular motors, 91, 97, 100, 103, 105,

106, 110

morphogenesis, 108, 168

multi-component systems, 28

nano-systems, 195, 205

Navier–Stokes’ equation, 62, 63, 160,

270

Nernst equation, 89

Neumann–Duhamel’s relation, 245

Newton’s cooling law, 65, 149, 176

Newton–Stokes’ law, 37, 199

non-equilibrium entropy, 180, 207, 257,

279, 289, 304

non-equilibrium temperature, 189, 190,

197, 250, 255, 257, 277, 291, 295,

301

Non-Fickian diﬀusion, 273, 274, 278

non-local eﬀects, 179, 192, 193, 205, 207,

219, 233

normal modes, 136, 137

normal stresses, 201

Ohm’s law, 38, 69, 71, 203

osmotic pressure, 83–85

Other sources, 128

overstability, 138

oxidative phosphorylation, 100, 103

parabolic diﬀerential equation, 60, 182

pattern formation, 136, 169, 172, 174

Peltier’s eﬀect, 73, 74, 76, 87

permeability coeﬃcient, 85, 87

phase speed, 183

phase transitions, 135, 143, 215

phenomenological coeﬃcients, 47, 49,

52, 53, 62, 64, 65, 68, 72, 81, 85,

87, 101, 102, 104, 111, 226, 231,

251, 294, 298

phenomenological equations, 48, 62, 72,

98, 100

phonons, 70, 182, 192–195, 210

photovoltaic cells, 114, 132

Piola–Kirchhoﬀ stress tensor, 246

Poisson brackets, 266, 275

polymers, 40, 63, 170, 180, 182, 205,

207, 208, 232, 274, 298

power, 2, 70, 73, 74, 76, 78, 88, 110, 113,

115, 117–121

power lasers, 69

Poynting–Thomson, 223, 234

Prandtl number, 147, 149

pressure tensor, 45, 61, 64, 67, 181, 193,

200, 201, 205, 206, 212, 213, 238,

270, 276

prey–predator system, 163

radiation, 35, 66, 125, 127, 128, 132, 238

rational extended thermodynamics, 200,

253, 254

rational thermodynamics, 179, 200, 219,

233, 237, 238, 249, 251, 253, 254,

259, 275–277, 304

Rayleigh number, 138, 147, 150, 151,

154, 155, 161, 174, 175, 177

Rayleigh–B´enard’s instability, 145, 147,

150, 152, 161, 174, 175

reaction–diﬀusion, 136, 208

Index 325

reciprocal relations, 49, 50, 56, 64, 69,

72, 87, 91, 96, 97, 99, 104, 231,

277, 279, 282, 285

reﬂection coeﬃcient, 86

refrigerators, 75, 114, 127, 131

relaxation time, 40, 95, 146, 170, 179,

180, 182, 184, 186, 188, 191, 192,

194, 195, 199, 200

reversible processes, 1, 5, 7, 9, 12, 34,

41–43, 46, 113, 115, 303

Reynolds number, 138

rheology, 201, 215, 223, 227, 251, 252,

261, 277, 305

roll pattern, 145

salt ﬁngers, 158, 169, 171

second law, 2, 3, 5, 8, 11, 12, 14, 18, 22,

23, 36, 39, 42

second sound, 183, 184, 192, 194, 208,

212

Seebeck’s eﬀect, 73, 74

self-adjoint problems, 138

self-organization, 91, 136, 163, 169, 174

semiconductors, 69, 74, 75, 79

shortwave, 125

silicon diode, 204

solar energy, 114, 129

Soret’s eﬀect, 82

sources, 79, 115, 117, 126, 127, 223, 238

stability of equilibrium, 1, 24, 26, 28, 30,

52, 60, 200, 270

stable steady states, 290

state, 1, 3–7, 10, 11, 13, 15, 16, 18

state variables, 3, 4, 6, 15, 24, 25, 40, 45,

49–51, 63, 64, 79, 180

steady states, 181, 202, 211, 288, 300

Stefan–Boltzmann law, 35, 131

stochastic noise, 289

stoichiometric coeﬃcients, 29, 97

Stokes’ law, 270

strain tensor, 217, 240, 243, 245, 246

stress tensor, 61, 155, 176, 223, 225, 226,

229, 231, 235, 238, 240, 242, 244,

246, 248, 252, 258, 275

subcritical instability, 142, 143, 157

supercritical stability, 140

superﬂuids, 63, 180, 182, 205, 208, 234

surface tension, 140, 142, 144, 154–156,

176

surface-tension, 158

suspensions, 180, 208, 227, 228, 231–234

symplectic manifold, 264

Taylor number, 138, 161, 175

Taylor’s instability, 158, 161, 162

telegrapher’s equation, 183, 184

temperature equation, 56, 259

thermal convection, 136, 143, 144, 158

thermal diﬀusion, 81, 82, 89, 146, 147,

228

thermal waves, 184, 191–193, 210, 211

thermocouples, 73, 74

thermodiﬀusion, 18, 38, 54, 69, 70

thermodynamic degrees of freedom, 16

thermodynamic forces, 46, 47, 52, 54,

64, 68, 69, 111, 179, 283, 287

thermodynamic potentials, 19, 21, 24,

27, 35, 135, 233, 249, 261

thermoelasticity, 243

thermoelectric conversion, 76

thermoelectric generators, 72, 76, 102

thermoelectricity, 38, 47, 54, 69, 73

Thermophoresis, 89

Third Law, 14

Thomson eﬀects, 38

time reversal, 43, 49, 226, 264, 269–271,

285

time-reversal, 284, 294

transcritical bifurcation, 142

transport equations, 37, 47, 61, 179, 180,

192, 205

triangular chemical, 91

Turing Structures, 167

Turing’s instability, 173

two temperatures, 73

ultraﬁltration coeﬃcient, 85

ultrasound propagation, 40, 64, 180, 199

upper-convected Maxwell model, 201

van’t Hoﬀ Relation, 31, 32

(extensive) variables, 13

variational principles, 51, 53, 68

velocity of reaction, 29, 92, 94, 110

viscoelasticity, 221

viscosity, 62, 63, 67, 68, 79, 105, 117,

140, 142, 146, 147, 158, 175, 199,

201, 212, 225, 234, 248, 251, 269,

273, 276, 286

viscous ﬂuids, 59, 181, 270

Volterra derivative, 265

wave equation, 183

Wien’s law of radiation, 125

wind energy, 114, 130

work, 2, 6–8, 10, 14, 21, 24, 34, 36, 37,

46, 54, 57, 92, 100

Zeroth Law, 5

Understanding Non-equilibrium Thermodynamics

G. Lebon • D. Jou • J. Casas-Vázquez

Understanding Non-equilibrium Thermodynamics

Foundations, Applications, Frontiers

123

Prof. Dr. David Jou Universitat Autònoma de Barcelona Dept. Fisica - Ediﬁci Cc Grup Fisica Estadistica Bellaterra 08193 Catalonia, Spain David.Jou@uab.es Prof. Dr. Georgy Lebon Université de Liège Dept. d’Astrophysique G´ ophysique et Oc´ anographie e e B5 Sart Tilman Liege 4000 Belgium g.lebon@ulg.ac.be

Prof. Dr. José Casas-Vázquez Universitat Autònoma de Barcelona Dept. Fisica - Ediﬁci Cc Grup Fisica Estadistica Bellaterra 08193 Catalonia, Spain Jose.Casas@uab.es

Cover image: Thermal radiation leaving the Earth, seen by the EOS-Terra satellite (NASA). Image from www.visibleearth.nasa.gov. Owner: NASA.

ISBN: 978-3-540-74251-7

e-ISBN: 978-3-540-74252-4

Library of Congress Control Number: 2007935452 c 2008 Springer-Verlag Berlin Heidelberg This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, speciﬁcally the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microﬁlm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable to prosecution under the German Copyright Law. The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a speciﬁc statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Cover design: WMX Design GmbH, Heidelberg Printed on acid-free paper 9 8 7 6 5 4 3 2 1 springer.com

Preface

Our time is characterized by an explosion of information and by an acceleration of knowledge. A book cannot compete with the huge amount of data available on the Web. However, to assimilate all this information, it is necessary to structure our knowledge in a useful conceptual framework. The purpose of the present work is to provide such a structure for students and researchers interested by the current state of the art of non-equilibrium thermodynamics. The main features of the book are a concise and critical presentation of the basic ideas, illustrated by a series of examples, selected not only for their pedagogical value but also for the perspectives oﬀered by recent technological advances. This book is aimed at students and researchers in physics, chemistry, engineering, material sciences, and biology. We have been guided by two apparently antagonistic objectives: generality and simplicity. To make the book accessible to a large audience of nonspecialists, we have decided about a simpliﬁed but rigorous presentation. Emphasis is put on the underlying physical background without sacriﬁcing mathematical rigour, the several formalisms being illustrated by a list of examples and problems. All over this work, we have been guided by the formula: “Get the more from the less”, with the purpose to make a maximum of people aware of a maximum of knowledge from a minimum of basic tools. Besides being an introductory text, our objective is to present an overview, as general as possible, of the more recent developments in non-equilibrium thermodynamics, especially beyond the local equilibrium description. This is partially a terra incognita, an unknown land, because basic concepts as temperature, entropy, and the validity of the second law become problematic beyond the local equilibrium hypothesis. The answers provided up to now must be considered as partial and provisional, but are nevertheless worth to be examined. Chapters 1 and 2 are introductory chapters in which the main concepts underlying equilibrium thermodynamics and classical non-equilibrium thermodynamics are stated. The basic notions are discussed with special emphasis on these needed later in this book. V

VI

Preface

Several applications of classical non-equilibrium thermodynamics are presented in Chaps. 3 and 4. These illustrations have not been chosen arbitrarily, but keeping in mind the perspectives opened by recent technological advancements. For instance, advances in material sciences have led to promising possibilities for thermoelectric devices; localized intense laser heating used to make easier the separation of molecules has contributed to a revival of interest in thermodiﬀusion; chemical reactions are of special interest in biology, in relation with their coupling with active transport across membranes and recent developments of molecular motors. The purpose of Chaps. 5 and 6 is to discuss two particular aspects of classical non-equilibrium thermodynamics which have been the subject of active research during the last decades. Chapter 5 is devoted to ﬁnite-time thermodynamics whose main concern is the competition between maximum eﬃciency and maximum power and its impact on economy and ecology. This classical subject is treated here in an updated form, taking into account the last technological possibilities and challenges, as well as some social concerns. Chapter 6 deals with instabilities and pattern formation; organized structures occur in closed and open systems as a consequence of ﬂuctuations growing far from equilibrium under the action of external forces. Patterns are observed in a multitude of our daily life experiences, like in hydrodynamics, biology, chemistry, electricity, material sciences, or geology. After introducing the mathematical theory of stability, several examples of ordered structures are analysed with a special attention to the celebrated B´nard cells. e Chapters 1–6 may provide a self-consistent basis for a graduate introductory course in non-equilibrium thermodynamics. In the remainder of the book, we go beyond the framework of the classical description and spend some time to address and compare the most recent developments in non-equilibrium thermodynamics. Chapters 7–11 will be of interest for students and researchers, who feel attracted by new scientiﬁc projects wherein they may be involved. This second part of the book may provide the basis for an advanced graduate or even postgraduate course on the several trends in contemporary thermodynamics. The coexistence of several schools in non-equilibrium thermodynamics is a reality; it is not a surprise in view of the complexity of most macroscopic systems and the fact that some basic notions as temperature and entropy are not univocally deﬁned outside equilibrium. To appreciate this form of multiculturalism in a positive sense, it is obviously necessary to know what are the foundations of these theories and to which extent they are related. A superﬁcial inspection reveals that some viewpoints are overlapping but none of them is rigorously equivalent to the other. A detailed and complete understanding of the relationship among the diverse schools turns out to be not an easy task. The ﬁrst diﬃculty stems from the fact that each approach is associated with a certain insight, we may even say an intuition or feeling that is sometimes rather diﬃcult to apprehend. Also some unavoidable diﬀerences in the terminology and the notation do not facilitate the communication. Another

results. 7). Moreover. it is our belief that a good comprehension of these diﬀerent versions allows for a better and more understandable comprehension of theories whose opportunity was not oﬀered to be discussed here. memory and non-local eﬀects. theories with internal variables (Chap. within the perspectives of improvement and uniﬁcation. Chapters 7–11 are autonomous and self-consistent. advantages. It could be surprising that the book is completely devoted to macroscopic and mesoscopic aspects and that microscopic theories have been widely omitted. At the end of each chapter is given a list of problems. it is highly recommended to browse through all the chapters to better apprehend the essence and the complementarity of the diverse theories. Comparison of diﬀerent viewpoints may be helpful for a deeper comprehension and a possible synthesis of the many faces of the theory.Preface VII factor that contributes to the diﬃculty to reaching a mutual comprehension is that the schools are not frozen in time: they evolve as a consequence of internal dynamics and by contact with others. It is clear that our choice is subjective: we have nevertheless been guided not only by the pedagogical aspect and/or the impact and universality of the diﬀerent formalisms. 9). One problem was the selection of the main representative ones among the wealth of thermodynamic formalisms. but also to stimulate his interest to solve concrete situations. we hope to convince the reader that macroscopic approaches. Notwithstanding. 7–11 are not only to get rid of the local equilibrium hypothesis. Some of these problems have been . 8). but also by the fact that we had to restrict ourselves. and mesoscopic approaches (Chap. like thermodynamics. it is highly desirable to include as many microscopic results as possible into the macroscopic framework. they have been structured in such a way that they can be read independently of each other and in arbitrary order. Here we have focused our attention on ﬁve of them: extended thermodynamics (Chap. rational thermodynamics (Chap.11). Hamiltonian formulation (Chap. Our goal is to contribute to a better understanding among the diﬀerent schools by discussing their main concepts. The aim is not only to allow the reader to check his understanding. 10). deserve a careful attention and are the seeds of the progress of knowledge. The common points shared by the theories presented in Chaps. However. which is the pillar of the classical theory. Such a comparative study is not found in other textbooks. In each of them. we have tried to save the particular spirit of each theory. Although statistical mechanics appears to be more fashionable than thermodynamics in the eyes of some people and the developments of microscopic methods are challenging. and limitations. we remain convinced that. with the purpose to account for the technological requirements of faster processes and more miniaturized devices. The reasons are that many excellent treatises have been written on microscopic theories and that we decided to keep the volume of the book to a reasonable ratio. but also to propose new phenomenological approaches involving non-linearities.

and of the Direcci´ General de Recerca of the Generalitat of o Catalonia. Jou. and which may be consulted for further investigation. March 2007 e G. J. for the discuse sions on these and related topics for more than 30 years. and in particular M. Li`ge-Bellaterra. Casas-V´zquez a . Lebon. Desaive (Li`ge University). under grants 2001 SGR 00186 and 2005 SGR 00087. Grmela (Montreal University). which are mentioned. Drs. P. We also appreciate our close collaborators for their help and stimulus in research and teaching. We also acknowledge the ﬁnantial support of the Direcci´n General de Investigaci´n of the o o Spanish Ministry of Education under grants BFM2003-06003 and FIS200612296-C02-01. We acknowledge many colleagues. Vicen¸ M´ndez and Vicente Ortega-Cejas deserve special gratitude c e for their help in the technical preparation of this book. D.C Dauby and Th. More technical and advanced parts are conﬁned in boxes and can be omitted during a ﬁrst reading.VIII Preface inspired by recent papers.

. . . . . . . . . .4. . . . . . . . 1. . . . . . . 1. . 1. 1. . . . . . . . . . . .3. . . . 1. . . . . 1. . . . . . . . . . . . . 1 1 3 5 5 6 8 14 14 15 16 16 17 18 19 20 21 24 24 27 27 29 30 31 32 34 34 IX . . . 1. . . . . . . . . .3 Stability of Chemical Equilibrium and Le Chatelier’s Principle . . . . .4 Gibbs’ Equation . . . . . . . . . . . . 1. . . . .6. . 1. . 1. . . . . .4. . . . 1. . . . . . . . . . 1. . . . . . . . . .1 Thermodynamic Potentials . . . .5 The Basic Problem of Equilibrium Thermodynamics . . . . . . . . . . . . . . . .1 Stability of Single Component Systems . . . . . . . . . . . . . . . . . . .6. . . . .4. . .2 Euler’s Relation . . . . . . . . . . . . . .3. . . . . . . . . . . . .4. . . . . .Contents 1 Equilibrium Thermodynamics: A Review . . . . . . . . . . . 1. . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . 1. . . . 1. . . . 1. . .7.2 Heat of Reaction and van’t Hoﬀ Relation . . . . . . . . . . . . . . . . . . . . . . .2 Scope and Deﬁnitions . . . . . . . .1 General Equilibrium Conditions . . . . . . . . . . . . 1. . . . 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Gibbs–Duhem’s Relation . . . . . . . . . . . . . . . . . . . . . . . . . 1. . 1. . . . .3. . . . . . . . . . . . . . .4 Some Deﬁnitions . . . . . . . .5 Legendre Transformations and Thermodynamic Potentials . . . 1. .1 Fundamental Relations and State Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7. . . . . .6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1. . . 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5. . . . . . . . . . . . . . .7 Equilibrium Chemical Thermodynamics . . . . . . .2 The First Law or Energy Balance . . . . 1. . . 1. . . . . . . .4. . .1 The Early History . . . . . . . . . .1 The Zeroth Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8 Final Comments . . . . . . . . .3 The Fundamental Laws . . . .2 Thermodynamic Potentials and Extremum Principles . .4 The Third Law . . . . . . . . 1.6 Stability of Equilibrium States .3 Stability Criterion of Multi-Component Mixtures . . .5. . . .2 Stability Conditions for the Other Thermodynamic Potentials . . . . . . . . . . . . . . . . . . . . .7. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9 Problems . . . . . . . .3 The Second Law . . . . . .

. . . . . .2 Matter Diﬀusion Under Isothermal and Isobaric Conditions . . . . . . . . 2. . . . . . . . . . . . .1. . . . . . . . . . .6 Problems . . . . . . . and Fluid Flows . . . . . . .1 Curzon–Ahlborn’s Model: Heat Losses . .1 Electrical Conduction . . . . . . . . . . . . . . 3. . . . . . . . . . . . . . . . . . 4. . . . . . . . 5. . . . . . . . . 5.3 Thermodiﬀusion: Coupling of Heat and Mass Transport . . . . .1. . .5 Autocatalytic Reactions and Diﬀusion: Morphogenesis . . . . .6. . . . . . . . . . . . . .4. . . . . . . . . . .2. . . . 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 38 39 41 44 50 51 54 54 59 60 63 65 69 70 72 72 76 79 83 83 85 87 91 92 96 96 97 100 102 108 109 113 114 115 120 122 3 4 5 . . . . . . . . . . . . 2. . . . . . . . . . . . . . 3. . . . . 2.6. . . .2 Cyclical Chemical Reactions and Onsager’s Reciprocal Relations . . . . . . . . . . . . . . . . . . . . .1 General Formalism . . .1 Phenomenological Laws . . . . . . . . . . . . . 3. 3. . . . . . . . . .1 Basic Concepts . . . . 4.2 Thermoelectric Eﬀects . . . . . . . . . . . . . .5 Problems . . . . . . . . . .1 Heat Conduction in a Rigid Body . . Chemical Reactions and Molecular Machines . . . . . . . . . . .1 Minimum Entropy Production Principle . . . . . . . . . . . 2. . . . . . . . .2 Economical and Ecological Constraints . . . . . . . . . . . . . . . . . . . .2. . . . . . . .7 Limitations of the Classical Theory of Irreversible Thermodynamics . . . . . . 4. Mass Transport. . . . . . . . . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .X Contents 2 Classical Irreversible Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . . . . . . 3. . . . . . . . . . . . . . . . 2. .6 Applications to Heat Conduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . . 3. . . . . . . . . . . . .3 Eﬃciency of Energy Transfer . . . . . . . . . . . . . . . . . . . 4. . . . . . . . . . . . .1 Entropy Production . . . . . . . . . . . . . . . . . .2. . . . . . . 2. . . . . .5 Stationary States . . . . . . . . . . . 4. . . . . . . . . . . . . . . . . . . . 3.6. . . . . . . . . .2 Eﬃciency of Thermoelectric Generators . . .2 Local Equilibrium Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Friction Losses . . . . . Coupled Transport Phenomena . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5. . 2. . . . . 4. . .1 One Single Chemical Reaction . . . . . . . . . . .3 Hydrodynamics . . . . . . . .4 General Theory . .4 Chemical Reactions and Mass Transport: Molecular Machines . . . . .3 Entropy Balance . . . . . . . . . . . . . . . . . . . 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. . . .1 The Finite-Time Carnot Cycle . . . . .5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . Finite-Time Thermodynamics . . . .8 Problems . . .2.4 Diﬀusion Through a Membrane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . . . . 5. . . . . . . .2 Phenomenological Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Coupled Chemical Reactions . . . . . . 3. . . . . . . . . . . . . . . . . 2. .

. . . . . . .2 The Rayleigh–B´nard’s Instability: e A Non-Linear Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7. . . . . . . . . . . . . . .2 Global Warming . 5. . .2 Patterns in Electricity . .1 Earth’s Energy Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 The Rayleigh–B´nard’s Instability: A Linear Theory .3. . . . . . . . . .1 Heat Conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6. . . . . . . . . . . . . . . . . . . . . . 6. . . . . 6 Instabilities and Pattern Formation . .5 Final Comments and Perspectives . . .3 Transformation of Solar Heat into Wind Motion . . .3. . . . .1 Temporal Organization in Spatially Homogeneous Systems . 6. . . . . . . . . . . . . . .3. . . . . . . . . . . . 6.5 Chemical Instabilities . . . . .1. . . .2 Spatial Organization in Spatially Heterogeneous Systems . . . . . . . . . . . . . . . . . . . . . . . . 7. . . . . . . . . . . . . . . . . . . . .3 B´nard–Marangoni’s Surface Tension-Driven e Instability . . . . . 6. . . . .3 Rheological Fluids .3 Spatio-Temporal Patterns in Heterogeneous Systems: Turing Structures . . . .2 Non-Linear Approaches . . . . . . . . . . . .4 Taylor’s Instability . . . . . . . . .6 Problems . . 6.1 Fourier’s vs. . . . . . . . . . . . . .5. . . . . . . . . .1. . . . . . . . . .6 Miscellaneous Examples of Pattern Formation . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5. . . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7. . . 6. . . . . . . . .1 The Linear Theory of Stability . . .3 Non-Local Terms: From Collision-Dominated Regime to Ballistic Regime . . . . . . .